Guidance in force

Public edition · one portal · open to everyone

AI Capability Governance and Implementation

One trusted route to discover, assess, implement, govern and get value from AI in public services. Governance follows what the AI can see, create and do, whichever supplier made it.

  • 76chapters
  • 12parts
  • 7working tools
  • 69tiered sources
  • 29signals
  • 62glossary terms

Four non-negotiable principles

One route from idea to lasting public value

  • Humans remain accountable. AI never grants authority a person does not already hold.
  • AI outputs are verified, not trusted by default. Every claim traces to a citable source.
  • Public value, equality and accessibility are built in — screened before delivery, not audited after.
  • Evidence, metadata and provenance are mandatory for anything material or public-facing.

Spotlight · new in 3.3.0

The agent governance playbook is now part of this portal

Chapters on criticality and tiers, registers and transparency records, oversight, cost and value, people and adoption, a live-pilot method and the delivery documents behind it, with one glossary and one contents drawer for everything.

Open the contents What was withheld

Start here

Pick the route that matches your job today

Task-first entry points, because people arrive with a decision to make — not an interest in our information architecture.

How much do you already know?

Choose a level, then a route below. Every route and task is marked with its level.

Routes by role

Common tasks

Not sure whether to use AI at all? Ask three questions first: is there a simpler non-AI way to meet the user need; can you explain the decision to the person it affects; and could you defend this use at a public inquiry? If any answer is no, stop and seek advice.

Contents · 76 chapters in 12 parts

Every part, every chapter

The agent governance playbook takes a team from “can I use AI for this?” to a governed, monitored and eventually retired AI agent. Open a part to see its chapters, or use the contents drawer.

Part 1 · 5 chaptersStart hereFrom Overview to Capability domains
  1. 1.1 Overview
  2. 1.2 Can I use AI for this?
  3. 1.3 How to use this guide
  4. 1.4 The playbook
  5. 1.5 Capability domains
Part 2 · 6 chaptersDecide and prioritiseFrom Evidence before assertion to Assurance tiers
  1. 2.1 Evidence before assertion
  2. 2.2 Use cases & evidence
  3. 2.3 Public sector case library
  4. 2.4 Classify & assess
  5. 2.5 Criticality & gates
  6. 2.6 Assurance tiers
Part 3 · 8 chaptersDesign and buildFrom Lifecycle & templates to Autonomy boundaries
  1. 3.1 Lifecycle & templates
  2. 3.2 Capability classes
  3. 3.3 Skill studio
  4. 3.4 Agent assurance X-ray
  5. 3.5 Connectors, MCP & integrations
  6. 3.6 Architecture & quality
  7. 3.7 Knowledge foundation
  8. 3.8 Autonomy boundaries
Part 4 · 8 chaptersAssure and governFrom Control plane & evidence to Failure & recovery
  1. 4.1 Control plane & evidence
  2. 4.2 Data & information governance
  3. 4.3 Architecture of trust
  4. 4.4 Governance operating model
  5. 4.5 Risks & mitigations
  6. 4.6 Registers & lifecycle
  7. 4.7 Internal ATRS registry
  8. 4.8 Failure & recovery
Part 5 · 6 chaptersOperate and overseeFrom Portfolio control tower to Observatory & barometer
  1. 5.1 Portfolio control tower
  2. 5.2 Monitoring & reassessment
  3. 5.3 Agent journey
  4. 5.4 Agent employment
  5. 5.5 Public service impact
  6. 5.6 Observatory & barometer
Part 6 · 4 chaptersFund and buyFrom Cost, value & investment to Dependency & exit
  1. 6.1 Cost, value & investment
  2. 6.2 FinOps & models
  3. 6.3 Commercial & suppliers
  4. 6.4 Dependency & exit
Part 7 · 11 chaptersPeople and adoptionFrom Multidisciplinary teams to Adoption & change model
  1. 7.1 Multidisciplinary teams
  2. 7.2 AI service team
  3. 7.3 Roles & pathways
  4. 7.4 Capability academy
  5. 7.5 Capability & maturity
  6. 7.6 Inclusive adoption
  7. 7.7 Adoption journey
  8. 7.8 Agent rollout
  9. 7.9 Community engagement
  10. 7.10 Operating model
  11. 7.11 Adoption & change model
Part 8 · 9 chaptersLive pilotFrom The pilot in one view to Decision & toolkit
  1. 8.1 The pilot in one view
  2. 8.2 What it tests
  3. 8.3 Candidate selection
  4. 8.4 Pilot roles
  5. 8.5 Assurance model
  6. 8.6 Journey & ceremonies
  7. 8.7 Communications
  8. 8.8 Success & feedback
  9. 8.9 Decision & toolkit
Part 9 · 8 chaptersProduct documentationFrom Product requirements to Gov.uk AI resources
  1. 9.1 Product requirements
  2. 9.2 Functional & technical spec
  3. 9.3 Solution architecture
  4. 9.4 High-level design
  5. 9.5 Low-level design
  6. 9.6 Theory of change
  7. 9.7 Personas & journeys
  8. 9.8 Gov.uk AI resources
Part 10 · 4 chaptersKnowledge and navigationFrom Thesaurus to Federated guidance service
  1. 10.1 Thesaurus
  2. 10.2 Taxonomies & ontology
  3. 10.3 Platforms & repositories
  4. 10.4 Federated guidance service
Part 11 · 4 chaptersDeploy for realFrom Service atlas to What this does not cover
  1. 11.1 Service atlas
  2. 11.2 Case generator
  3. 11.3 Worked examples
  4. 11.4 What this does not cover
Part 12 · 3 chaptersReferenceFrom UK nations to Pace & parity
  1. 12.1 UK nations
  2. 12.2 International
  3. 12.3 Pace & parity

Part 1 of 12 · 5 chapters

Start here

Part 1 · Start hereChapter 1.1 · 1 of 76

AI Capability Governance and Implementation

The guidance for deploying AI agents in the public sector, sequenced into one route — and the operating capability that turns experimentation into governed services. Crown-copyright sources are the government's own; every illustrative figure is labelled.

Start here

Eight stages from mandate to governed operation. The last stage feeds the third: every review either re-authorises the agent, changes it, or ends it.

Stage 8 returns to stage 3 — the loop does not close

Usage and cost Who uses it, how often, what it spends

Accuracy and drift Evaluation results against the baseline

Risk and incidents New risks, near misses, escalations

Value realised Benefits claimed against benefits evidenced

Review or retire Re-authorise, improve, restrict or decommission

Monitoring and human oversight are continuous duties held by the named owner, not a phase that ends at go-live.

Where this guide sits

The hierarchy of AI-related guidance a public-sector team works inside. This guide is a proposal: it interprets and sequences the guidance above it, and it does not replace or override any of it.

International frameworks

OECD AI Principles, EU AI Act risk tiers, ISO/IEC 42001 management standards.

UK statute and cross-government policy

Data Protection Act 2018 and UK GDPR, Equality Act 2010, Procurement Act 2023, the AI Opportunities Action Plan.

Central guidance in force

AI Playbook for the UK Government, Service Standard, Technology Code of Practice, ATRS, Data Ethics Framework, NCSC secure-AI guidelines.

Departmental policy and standards

Organisational AI policy, data classification, security standards, delegated authority and spend controls.

This implementation guide

AI Capability Governance and Implementation — 69 chapters, assessment tools and register schemas that turn the guidance above into delivery.

Proposal

Team-level artefacts

Agent register entries, ATRS records, assurance evidence, instruction blocks and cost models.

Multidisciplinary from the outset

Every source in this app lands on the same point. The AI Playbook makes skills and expertise a principle in its own right. The Service Manual builds the multidisciplinary team into the Service Standard. NCSC wrote its secure-AI guidelines for data scientists, developers, managers, decision-makers and risk owners together. The roster is in Teams; the readiness barometer is in the Observatory.

Back to the contents · Suggest a correction (opens GitHub)

Part 1 · Start hereChapter 1.2 · 2 of 76

Responsible AI · start here

What do you want to do?

Start from your task. This part of the playbook routes you to the proportionate tool, template, assessment or support. Governance follows what the AI can see, create and do, not which company made it.

Quick route selector · five questions

Can I use AI for this?

1. Would the output decide, or directly change, someone’s rights, entitlements, safety or access to a service?
2. Will you use personal data or information marked OFFICIAL-SENSITIVE or above?
3. Can the AI take actions, run on a schedule or work through several steps without you?
4. Is there an approved tool in your organisation's catalogue that already does this?
5. Could a conventional rule, form or automation do the job reliably?

Your route

Answer the five questions

The route appears when every question has an answer. “Not sure” is a valid answer.

A first sort, not an approval. Any route other than an approved tool leads to a registered use case.

Six things to know first

AI concepts and limitations

Generative AI predicts plausible output. It can be fluent and wrong at once, so outputs need checking in proportion to what depends on them.

Models and providers

A product can use several models from several providers, sometimes chosen automatically. The model can change without the product name changing.

Enterprise grounding

Grounding feeds organisational content to the model. Whatever the index can reach, the AI can surface, including titles, filenames and metadata.

Agents, tools and actions

An agent plans steps and uses tools to act: send, update, create or call other agents. Acting changes the risk boundary.

Human oversight

Oversight works only if the reviewer has the competence, the evidence and the authority to challenge, override or stop.

Deployment route

The same model reached directly, through a cloud platform or inside another supplier’s product is a different governance object each time.

For everyone in the organisation

All colleaguesFind guidance, approved tools and practical help

Leaders and managersMake informed decisions, manage risk and value

Risk, security and complianceAssure, monitor and advise

IT and digitalEnable, integrate and manage

Data and informationEnsure data is ready and well managed

Commercial and procurementManage suppliers, licences and costs

Accessibility and inclusionSupport all users and workplace adjustments

Policy, strategy and transformationDrive adoption, value and continuous improvement

Enabled by common services

Real-world outcomes · turning responsible AI into a stronger public sector

Safer services for citizens
Greater value for taxpayers
More efficient working
More inclusive for all
A more innovative public sector
Ideas → responsible implementation → trusted operation → lasting public value

Govern the capability and behaviour first

Microsoft Copilot, Anthropic Claude, GitHub Copilot, Gemini and future tools sit as product overlays beneath one common control model. Role routes for every audience are in How to use this guide.

Get help

Using an approved tool

Your IT service desk, then the product’s support route.

Assessing a use case

The AI assurance team owns screening and the decision register.

An incident or concern

Report through the incident route on the Operate page. Treat data exposure as a security incident.

Accessibility or adjustments

Your line manager and workplace adjustments service.

Named contacts are set by each organisation when the guidance is deployed.

Standards and policy

AI policy

Your organisation’s AI policy and acceptable use.

Government guidance

The AI Playbook for the UK Government.

AI Playbook on GOV.UK ↗
Data protection

UK GDPR, DPIAs and international transfers.

ICO guidance on AI ↗
Security

Threat modelling, secure configuration and incident response.

Records

Retention and disposal of prompts, outputs, transcripts and logs.

Accessibility

WCAG 2.2 AA for AI interfaces and outputs.

WCAG 2.2 ↗
Procurement

Supplier assurance, subprocessors, exit and service commitments.

Transparency

Algorithmic Transparency Recording Standard for public-facing tools.

ATRS hub on GOV.UK ↗
Templates

Twelve core artefacts and seven implementation artefacts.

Product guidance

Overlays for each supplier and deployment route.

Back to the contents · Suggest a correction (opens GitHub)

Part 1 · Start hereChapter 1.3 · 3 of 76

Start here · orientation

How to use this guide

87 chapters in 12 parts. Nobody needs all of it. Pick a reading mode, then a route.

Front to back

12 parts in delivery order: decide, build, assure, operate, fund, the people and pilot material, then the deployment part and the reference. Roughly a day's read.

By role route

10 routes below, each an ordered set of chapters for one job. Twenty to forty minutes.

By task

Use the launchpad or search: rate readiness, tier an agent, price a job, build an instruction block, check exit readiness.

Routes by role

Each route is an ordered set of chapters. Click any step to jump there.

Data and information

Get data ready, keep metadata and indexing honest, govern connectors and treat AI outputs as records.

All colleagues

Find an approved tool, get a route for your task, and know where to get help.

Accountable leader or SRO

Decide where AI is worth the risk and the spend, and be able to answer for it.

Service owner

Adopt agents without destabilising a service people depend on.

Product or delivery manager

Frame the problem, prove the value, keep the team moving.

AI engineer or technical owner

Build well-grounded agents and prove they work.

Assurance, ethics or audit

Apply proportionate control and see the evidence behind it.

Commercial, finance or FinOps

Fund adoption honestly and keep exit possible.

People, learning and adoption

Build capability that outlasts the pilot.

Pilot participant

Use the guide during a real deployment and record what happens.

What each part is for

1 · Start here

What the guidance says, what this guide adds, and the ten capability domains it all rests on.

2 · Decide and prioritise

Whether to do it at all: evidence quality, use-case selection, criticality and the assurance tier that follows.

3 · Design and build

Capability class — generative, agentic or declarative — then instructions, architecture, grounding and the autonomy boundary.

4 · Assure and govern

The trust cycle, the operating model, risks and mitigations, registers, transparency records and recovery.

5 · Operate and oversee

Life after approval: the portfolio view, the agent lifecycle, public-service impact and readiness tracking.

6 · Fund and buy

The full investment line, unit economics, supplier requirements and exit.

7 · People and adoption

Teams, roles, pathways, inclusion and the organisational adoption programme.

8 · Live pilot

How this guide is being tested in a real deployment, and what would justify a permanent toolkit.

9 · Product documentation

Requirements, architecture, design, theory of change and the research base for the guide itself.

10 · Knowledge and navigation

Glossary, thesaurus, generated knowledge graph, taxonomies, ontology, site map, guided tours and platform links.

11 · Deploy for real

The part that turns the guide into a deployment: fifty-four real services placed on volume and criticality, a generator that derives the whole governance set for any case, four agents carried end to end — one in each quadrant — and an explicit list of what this guide still does not cover.

12 · Reference

Devolved and international comparators, and every source behind the content.

How to read the labels

Verified Present in a surfaced source.

Interpretation A conclusion drawn across sources.

Proposal Recommended future state, not adopted policy.

Illustrative Looks like data, carries no measured source.

What this guide is not

It does not override departmental policy, professional advice or formal approvals.

It is not the register of record for any live deployment.

It holds no personal, case or OFFICIAL-SENSITIVE data; assessments stay on your device.

It cannot issue an approval — the accountable owner does that.

Back to the contents · Suggest a correction (opens GitHub)

Part 1 · Start hereChapter 1.4 · 4 of 76

The playbook

Ten principles, one life cycle

The AI Playbook for the UK Government (opens an external website) (GDS, 10 February 2025) supersedes the Generative AI Framework and covers all AI: machine learning, NLP, computer vision, generative and agentic systems. The ten principles are paraphrased below; open each for its first moves.

The deployment life cycle

Playbook principle 5: manage the whole life cycle, including the unglamorous parts. Each stage has a gate; nothing moves right without passing it.

Stage 1

Discover
  • Frame the problem and the user need: is AI the right tool at all?
  • Triage risk: personal data, decisions about people, public-facing?

Assemble the multidisciplinary team; screen for DPIA and equality impacts

Gate · A named SRO and a written problem statement

Stage 2

Design
  • Decide buy vs build; source and assess the data
  • Threat-model with security; place the human-oversight points

Legal, ethics and commercial shape the approach, not the finished thing

Gate · DPIA drafted; risks owned; oversight designed in

Stage 3

Build & evaluate
  • Develop securely; verify the model and data supply chain
  • Evaluate accuracy, bias and robustness against agreed baselines
  • Red-team it: prompt injection, misuse, edge cases

Gate · Evaluation results reviewed by the whole team

Stage 4

Deploy
  • Pass assurance gates; publish the ATRS record
  • Switch on human review and incident response
  • Tell users they are interacting with AI

Gate · Assurance sign-off; transparency record live

Stage 5

Operate & retire
  • Monitor drift, bias and hallucination in production
  • Re-evaluate on every model or data update
  • Plan decommissioning and knowledge handover

Gate · Monitoring dashboards live; exit plan on file

Back to the contents · Suggest a correction (opens GitHub)

Part 1 · Start hereChapter 1.5 · 5 of 76

The domains

Five lenses on every deployment

Cyber, data, digital, technology, and the legal-ethical frame around them. Each domain has its own authoritative guidance; a deployment is ready only when all five say so. That is why the team is multidisciplinary.

Cyber

Cyber security

NCSC secure AI guidelines
NCSC machine learning principles
Secure by Design (UK Government)

NCSC Guidelines for secure AI system development cover secure design, development, deployment and operation. Agreed with CISA and 21 international agencies; aimed at any system that uses AI, built or bought.

Secure by Design and the Government Cyber Security Standard apply to AI services like any other, plus AI-specific threat modelling: prompt injection, data poisoning, model extraction, over-permissioned agents.

NCSC's principles for the security of machine learning translate the threats into controls. Walk before you run with agentic AI.

Data

Data & information

Data Ethics Framework
ICO: AI and data protection
ATRS hub

The Data Ethics Framework guides appropriate, responsible data use across the public sector: transparency, accountability, fairness.

UK GDPR and the DPA 2018 require a DPIA before high-risk processing. The ICO's AI guidance, with a statutory AI code of practice in development, is the regulator's view of what good looks like. Cross-border processing (for example a US-processed model behind a UK product) needs a Transfer Risk Assessment as well as SCCs.

The Algorithmic Transparency Recording Standard is mandatory for central departments and arms-length bodies: a published record for every in-scope algorithmic tool.

Model quality is data quality: govern provenance, representativeness and retention for training and grounding data alike — retrieval systems inherit the permissions, quality and currency of the estate.

Digital

Digital & service

Service Standard
AI Playbook for the UK Government
AI Knowledge Hub

The Service Standard and Service Manual set the operating model: start with user needs, work in multidisciplinary teams, iterate in the open. AI services are still services.

The AI Playbook and the AI Knowledge Hub supply the patterns, case studies and communities; the PM's AI Exemplars programme is the test-and-learn engine for scaling what works.

Accessibility is law, not polish: WCAG 2.2 AA under the Public Sector Bodies Accessibility Regulations 2018. AI interfaces (chat, voice, summaries) need their own accessibility evidence.

Technology

Technology & architecture

Technology Code of Practice
AI Opportunities Action Plan
Action Plan delivery progress

The Technology Code of Practice governs the technology choices: cloud first, interoperability, open standards, sustainability, exit.

Compute and infrastructure are national policy: AI Growth Zones and a compute roadmap from 21 ExaFLOPs (2025) towards 420 by 2030 under the AI Opportunities Action Plan.

Buy well: use Crown Commercial Service AI routes, insist on evaluation access and model-change notice in contracts, and design the exit before signing the entry.

Legal & ethics

Legal, ethics & society

Explaining decisions made with AI
Data Ethics Framework
Equality Act 2010 guidance

The Equality Act 2010 and the Public Sector Equality Duty apply to automated and AI-assisted decisions. Public-law duties (fairness, rationality, giving reasons) still bind when a model is in the loop.

Explainability has a manual: 'Explaining decisions made with AI', by the ICO and The Alan Turing Institute — and explanations shown to people should reference functional needs, never a diagnosis or label.

Public trust is earned in the open: engage affected people early, say clearly where AI is used, and give a route to challenge and redress.

Back to the contents · Suggest a correction (opens GitHub)

Part 2 of 12 · 6 chapters

Decide and prioritise

Part 2 · Decide and prioritiseChapter 2.1 · 6 of 76

The evidence

Same product. Four UK evaluations. Four answers.

What varies is method, not the tool. Uncontrolled self-report gives the highest figures; comparison-group econometrics gives mid-range; counterfactual-free evaluation of a short pilot gives the lowest. Role mix is the second driver.

26

minutes / day

Cross-government, GDS

20,000 staff, self-reported

60

minutes / week

HMRC, randomised cohort

3,000 licences

19

minutes / day

DWP, comparison group

3,549 staff, econometric

0

robust evidence

DBT, no counterfactual

1,000 licences

Never average these numbers. The spread is evidence about evaluation design. Sources: GDS findings report (Jun 2025); HMRC phase 3; DWP evaluation (Jan 2026); DBT evaluation (Aug 2025).

Four things block the move to governed scale

Unremediated permissions

AI surfaces what a user may already reach. Permission drift becomes an incident on day one.

Makes scaling UNSAFE

No agent inventory

The transparency standard records algorithmic tools. It is not an operational register of agents.

Makes scaling UNGOVERNABLE

Unpredictable consumption cost

Credits are billed per feature, not per conversation. A builder's design choice sets a budget holder's cost.

Makes scaling UNBUDGETABLE

Unfunded human effort

Workflow redesign, local support and output verification happen but rarely appear as funded lines.

Makes scaling UNSUSTAINABLE

Activity is running ahead of visibility

Between deployment and disclosure sits a gap no current UK instrument closes. The ATRS is a transparency mechanism, not an inventory: it does not capture team-shared agents, orphaned agents, data-source approval, consumption cost or retirement.

21%

of bodies responding to the NAO had an AI strategy

NAO · Mar 2024

74

use cases deployed, reported at the same date

NAO · Mar 2024

33 → 125

transparency records: Jan 2025 → end-2025

PAC · OECD.AI

28%

of central government IT is legacy

PAC · Mar 2025

Coding agents: two findings, both real

A finding about where to deploy, not whether.

1 hr

saved per day — GDS cross-government trial, 50 organisations

72% said good value; 58% would not return to working without it

19%

slower — METR randomised trial: experienced developers, mature code

20% felt faster — the same developers, in the same trial

Benefit is largest on well-scoped, greenfield and boilerplate work; smallest or negative on mature, complex codebases. Developer self-assessment is unreliable in both directions. Join the AI Engineering Lab (OCTO, GDS) before procuring independently — the evidence you generate is reusable.

What the enthusiastic version leaves out

Contrary evidence, recorded deliberately.

17%

of cross-government trial users noticed no clear time savings (GDS, Jun 2025)

7%

of Australian participants reported the tool added time (DTA, Oct 2024)

61%

of Australian managers could not confidently identify AI outputs (DTA, Oct 2024)

72%

of councils were not ready on workforce skills and expertise (LGA, Jun 2025)

Canada names a further risk no UK instrument addresses: skill atrophy and automation drift. Skill maintenance is a prerequisite for meaningful human oversight, not an optional extra (TBS Canada, May 2026).

A caution on AI-produced research

Twenty sources in one AI research draft. What checking found:

7

verified precisely

6

verified in substance

4

not verified as cited

1

could not be located

None of the unverifiable entries was obvious: each was correctly formatted, plausibly dated, attributed to a real institution. Treat AI-produced research as a hypothesis-generating artefact, never an evidence base — and keep the verification log with the output.

Evidence labels used across this app

Verified fact — supported by official, regulatory, audit or primary sources

Organisation-reported — reported by the organisation itself, not independently measured

Supplier claim — authoritative for product terms; interested-party evidence for value

Interpretation — analytical judgement drawn from cited evidence

Recommendation — action proposed, not established practice

Unresolved — evidence incomplete, conflicting or arrangement-specific

Back to the contents · Suggest a correction (opens GitHub)

Part 2 · Decide and prioritiseChapter 2.2 · 7 of 76

Use cases & evidence

Where AI creates value across government and public services

Eighteen categories, from productivity tools to enterprise agents — supporting people, processes and public outcomes. Use AI and agents responsibly to enhance the work of public servants and improve outcomes for citizens.

1. Knowledge retrieval & insight Find, summarise and synthesise information from trusted sources

2. Content & correspondence Draft, edit and refine high-quality content and communications

3. Contact centres & helpdesks Support staff and citizens with accurate, consistent responses

4. Casework & service operations Streamline case handling, triage and workflow automation

5. Citizen services & engagement Deliver simpler, faster and more accessible services

6. Meetings & collaboration Capture, summarise and turn meetings into actions and outcomes

7. Data analysis & reporting Analyse data, create summaries and produce management information

8. Software engineering & IT operations Accelerate development, testing and secure operations

9. HR & people management Improve people processes and the employee experience

10. Finance & procurement Enhance financial planning, analysis and supplier management

11. Education, training & skills Personalise learning and improve knowledge retention

12. Programme & project management Plan, track and report with intelligent support

13. Health & care Support clinical and administrative workflows safely and securely

14. Policing, justice & public safety Improve investigations, evidence handling and public protection

15. Regulatory & compliance Monitor compliance and produce regulatory outputs

16. Policy & strategy development Scenario analysis, option generation and evidence synthesis

17. Communications & media Create campaigns, content and prepare responses

18. Research & horizon scanning Scan trends, research and emerging opportunities

Cross-cutting capabilities — enablers for all use cases

Secure access to trusted data & knowledge AI agents & orchestration (autonomy with oversight) Human oversight & accountability Privacy, security & data protection Quality assurance & model evaluation Observability, monitoring & auditability Integration with systems & workflows Change management & adoption FinOps & cost management

Foundational enablers

Data quality & governance Information management Access management Standards & interoperability Skills & culture Ethics & responsible AI

The use-case evidence card

A consistent way to describe, evaluate and share AI use-case evidence: what the capability does, who it helps, how it works, the value it delivers, and the evidence that supports it. Evidence first. Value always. Capture consistently, share openly, decide confidently.

1. Basic information Use case name · Short description · Unique ID · Date created · Last updated · Author / owner

2. Problem & purpose Problem / opportunity · Purpose & objectives · Success criteria

3. Users & context Primary user groups · Who benefits · Operating context / environment · Geography / organisation

4. Solution & capability AI capability / agent type · Key features / functions · How it works (high level) · Human oversight points

5. Data & information Key data sources · Type of data · Data sensitivity · Information governance

6. Implementation & integration Platform / technology · Systems / tools integrated · Deployment model · Date went live / expected · Resource requirements

7. Outcomes & value Quantitative outcomes (measured) · Qualitative outcomes (observed) · Value delivered (how & for whom)

8. Costs & financials Cost model (licence, usage, etc.) · Estimated / actual cost · Cost per user / transaction · Ongoing cost considerations · Funding source / owner

9. Risks & assurance Key risks · Mitigations in place · Assurance / evaluation approach

10. Evidence & validation Evidence summary · Evidence sources · What evidence is available (performance results, user feedback, evaluation report, audit / review, other)

11. Limitations & lessons Known limitations / constraints · Lessons learned / recommendations

12. Transferability & reuse Potential for reuse (high / medium / low) · Reusable artefacts available · Transferability notes

13. Lifecycle & review Lifecycle stage (ideation, pilot, live, scaled, retired) · Review frequency · Next review date · Sunset / retirement plan

Evidence strength
  • Strong — Independently validated
  • Medium — Reported results with some validation
  • Emerging — Pilot / early indications
  • Weak — Unverified claims
  • Unknown — Insufficient evidence
Evidence status

Live / in use Pilot / trial Planned Retired / stopped Unknown

Worked, failed and unproven

Sharing the full picture — successes, failures and emerging ideas — so we learn faster and build smarter. Transparent evidence helps the public sector avoid repeating mistakes and scale what works. The cases below are illustrative.

Worked — repeatable success

Evidence shows value, quality and user benefit.

Case: ACAS 'Haste' helpline knowledge assistant

Helps advisers quickly find accurate, impartial information during calls.

What worked
  • Grounded in authoritative, approved and consistent formats
  • Tested with 1,000 differently worded questions
  • High accuracy and consistency across impartiality and bias checks
  • Clear roles, assistant, not decision maker
  • Positive adviser feedback and time savings
Outcomes
  • Faster information retrieval
  • Improved accuracy and consistency
  • Better adviser experience
  • Scalable to wider teams

Failed — learned & corrected

Initial attempt did not deliver — lessons captured and applied.

Case: AI-assisted NHS document processing

Automated extraction from clinical documents produced high initial error rate.

What went wrong
  • Poor data quality and inconsistent document formats
  • Incomplete metadata and missing context
  • Model hallucinations and incorrect extractions
  • Insufficient validation and no independent review loop
  • Rushed from prototype to wider use
What we did next
  • Reduced manual re-keying and error tracking
  • Good accuracy in structured readings
  • Went back with clear agendas and permissions
  • User trust and change impact still being assessed
  • Scalability and cost model not yet validated

Unproven — promising, not yet validated

Early potential identified — evidence still limited.

Case: Agent automatically summarises meeting action agent

Summarises meeting decisions, actions and sends updates to owners.

What we know so far
  • Long-term testing across teams and meeting types
  • Long-term accuracy and retention unknown
  • Cost at scale and return on investment not established
  • Governance and data sensitivity assessments in progress
Next steps

Extended pilot, independent evaluation, costing and governance review before scaling

  • Data quality and validation are critical
  • Human review and iterative testing prevent scale of error
How we succeed together

Share openly, including failures Ground in authoritative sources Test with real users Measure impact and value Capture lessons and publish patterns Reuse and continuously improve

Four proven delivery patterns

Each pattern is a pipeline with human oversight built in — not a single tool. Use them as the starting shape for a new use case.

Meeting intelligence pipeline

Transforming conversations into decisions, actions and organisational memory

1. Meeting planning 2. Consent & capture 3. Transcript creation 4. AI processing & summarisation 5. Decisions & actions 6. Risks & insights 7. Distribution & integration 8. Organisational memory

Authoritative knowledge assistant

Delivering trusted, accurate answers grounded in authoritative sources, with human expertise in the loop

1. Define need 2. Retrieve 3. Evaluate & ground 4. Generate response 5. Human review 6. Deliver answer 7. Feedback & learn 8. Maintain sources

AI-assisted engineering lifecycle

Secure. Quality. Production-ready. Human in control.

1. Discover & plan 2. Architect & design 3. Build & code 4. Test & validate 5. Review & assure 6. Deploy & release 7. Operate & monitor

AI-assisted programme & portfolio management

Smarter oversight. Better decisions. Greater impact.

1. Capture & ingest 2. Analyse & understand 3. Plan & optimise 4. Monitor & track 5. Decide & recommend 6. Collaborate & execute 7. Learn & improve

Back to the contents · Suggest a correction (opens GitHub)

Part 2 · Decide and prioritiseChapter 2.3 · 8 of 76

Decide and prioritise · public sector case library

What central, local and devolved government has already done

Sourced cases from central government, councils, the Scottish, Welsh and Northern Ireland governments and health bodies. Each carries its source and an evidence grade. Supplier-published figures stay graded as reported.

Central government · Generative Corroborated

Humphrey: Consult, Parlex, Minute, Redbox and Lex

DSIT · i.AI, now part of GDS

A package of AI tools for civil servants: consultation analysis, parliamentary search, meeting transcription, document summarisation and legal research.

Governance lesson: Consult was tested on past consultations before live use.

Source: UKAuthority ↗

Central government · Generative Verified

Extract: digitising planning documents

i.AI with MHCLG

Extract digitises planning documents. Early testing completed a one-to-two-hour manual task in under a minute.

Governance lesson: Check extraction against the source before it feeds planning decisions.

Source: Roadmap for modern digital government ↗

Devolved: Scotland · Generative Verified

Consult tested live on a Scottish Government consultation

i.AI with the Scottish Government

The first live test of Consult used a Scottish Government consultation, followed by several other government consultations.

Governance lesson: Compare AI theming with human analysts before relying on it.

Source: Roadmap for modern digital government ↗

Local government · Generative Verified

Minute trialled in frontline local services

i.AI

Minute, a transcription and summarisation tool, was trialled in frontline local services in July 2025.

Governance lesson: Settle consent and retention for recordings before a trial starts.

Source: Roadmap for modern digital government ↗

Local government · Generative Verified

Magic Notes in adult social care

Kingston Council with Beam

Piloted AI recording and summarising of assessments to free social workers for direct care.

Governance lesson: A DPIA examined data flows, storage and access. The privacy notice was updated and practitioners get consent to record.

Source: Local Government Association ↗

Local government · Generative Reported

Magic Notes scaled to over 1,000 staff

Kent County Council

After 29 practitioners piloted the tool, more than 1,000 frontline staff use it. An evaluation reported seven hours a week saved on average.

Governance lesson: Figures are supplier-published. Validate locally before using them in a business case.

Source: techUK ↗

Local government · Generative Verified

Service-level privacy notice for AI transcription

Hartlepool Borough Council

A privacy notice for Magic Notes in adult social care sets out purposes, legal basis and rights.

Governance lesson: A published, service-level privacy notice is a reusable transparency artefact.

Source: Hartlepool Borough Council ↗

Local government · Generative Corroborated

AI recording in adult social care

Swindon Borough Council

Piloted AI to improve the quality and speed of recording assessment conversations.

Governance lesson: Evaluate quality of the record, not only time saved.

Source: Socitm ↗

Local government · Declarative Verified

Data & Analytics Board for data and AI governance

Belfast City Council

Set up a Data & Analytics Board to oversee data and AI governance, alongside a council-wide generative AI policy and staff training.

Governance lesson: Data and AI governance sit under one board.

Source: Belfast City Council ↗

Devolved: Scotland · Agentic Verified

Scotland’s AI Strategy 2026–2031

Scottish Government

Published 20 March 2026. An eight-layer “AI Stack” and a data matchmaking pilot for trusted public-sector datasets by 2027.

Governance lesson: Public-sector AI use is made visible through the Scottish AI Register.

Source: gov.scot ↗

Devolved: Scotland · Generative Verified

AI guidance on the Digital Scotland Service Manual

Scottish Government Public Sector AI Team

A suite of AI guidance and resources published on the Digital Scotland Service Manual in August 2026.

Governance lesson: Guidance lives in the service manual teams already use.

Source: Digital Scotland blog ↗

Devolved: Wales · Generative Verified

AI Cymru: the AI Plan for Wales

Welsh Government

AI adoption across health, education and local government, with a commitment that AI works equally well in Welsh and English.

Governance lesson: Public bodies assess risks to vulnerable groups and consider offline users.

Source: gov.wales ↗

Devolved: Wales · Generative Reported

Office for AI finds duplicated AI work

Welsh Government Office for AI

A baselining exercise found public bodies duplicating AI work. The office now coordinates adoption and says not every pilot should go live.

Governance lesson: Duplication is a governance failure; a shared register and reuse check fix it.

Source: Resultsense via UKAuthority ↗

Devolved: Wales · Agentic Reported

Senedd research on generative AI and agents

Senedd Research

A briefing on generative AI in Welsh public services covers inaccurate outputs, discrimination, privacy and the added security and accountability challenges of AI agents.

Governance lesson: Agent autonomy brings separate security and accountability questions.

Source: Think Digital Partners ↗

Health · Generative Reported

AI scribing in children’s neurodevelopmental services

Cwm Taf Morgannwg University Health Board with CDPS

A pilot assessing whether AI scribing reduces clinicians’ administrative workload.

Governance lesson: Pilot and evaluate before scaling clinical documentation tools.

Source: BusinessIn Wales ↗

Devolved: Northern Ireland · Agentic Verified

Draft Northern Ireland AI Strategy

The Executive Office

Consultation from 12 August to 7 October 2026. Eight principles include human oversight, accountability and redress, and data governance.

Governance lesson: Proposes human oversight teams, a cross-sector AI governance board and inventories of AI projects.

Source: Executive Office ↗

International comparator · Generative Corroborated

Guidelines for AI in the public service

Government of Ireland

Advises against free generative AI tools because information given to them could be used for training.

Governance lesson: A clear position on consumer tools reduces shadow AI.

Source: RTÉ ↗

17 cases in the library. Gathered from public sources on 19 October 2026. Re-check each source before citing it.

Back to the contents · Suggest a correction (opens GitHub)

Part 2 · Decide and prioritiseChapter 2.4 · 9 of 76

Responsible AI · assess a use case

Classify first, then answer only what your tier needs

As AI can see more, do more and act more independently, the assurance required rises. Criticality & gates scores the service impact; this chapter sets the route from what the AI itself can do.

Class

1. Ask

  • Answer and assist
  • Provides information, answers or summaries based on a prompt.
  • Example: Ask about a policy or summarise a document.
2. Create
  • Generate content
  • Creates new content, analysis, images or code.
  • Example: Draft a briefing, a presentation or code.
3. Connect

Use organisational information

Retrieves and works with data from organisational sources and external systems.

Example: Search SharePoint or analyse internal data.

4. Act
  • Take actions
  • Makes changes in systems, files or records, or triggers workflows.
  • Example: Create a record, send an email or start a process.
5. Delegate
  • Complete multi-step work
  • Plans and performs several steps with tools and limited supervision.
  • Example: Research, analyse, write a report and share it.
6. Persist
  • Operate proactively
  • Runs autonomously or on a schedule, monitoring and acting over time.
  • Example: Monitor changes, raise tickets or act on triggers.
Information access
  • User-provided only
  • User-provided only
  • Organisational sources
  • Organisational and system
  • Multiple sources and tools
  • Ongoing and broad
System access
  • None
  • None
  • Read (typically)
  • Read and write
  • Multiple systems
  • Multiple systems, scheduled
Autonomy
  • User led
  • User led
  • User led
  • User led, with action
  • Partial autonomy
  • High autonomy
Persistence
  • Single interaction
  • Single interaction
  • Single or repeated use
  • Repeated use
  • Multi-step tasks
  • Continuous or scheduled
Human oversight
  • Review output
  • Review output
  • Review output and sources
  • Review before or after action
  • Ongoing oversight
  • Defined oversight and monitoring
Assurance depth
  • Lightweight
  • Lightweight
  • Standard
  • Enhanced
  • High
  • Specialist

Generative, declarative and agentic: the three capability types

The capability classes say what the AI can do; the capability type says how it is built. Both set the assurance.

goes deeper on choosing between them.

Generative

Creates content in response to a prompt: text, summaries, images or code. A person decides what happens next.

Declarative

Configured, not coded. Purpose, instructions, knowledge and permitted actions are declared; the platform supplies the model and orchestration.

Agentic

Plans and takes actions across tools under delegated authority, choosing its own path to a goal, sometimes on a schedule.

ClassUsual capability typeMinimum assurance
AskGenerativeReview the output.
CreateGenerativeReview the output; label AI involvement.
ConnectGenerative, grounded · or Declarative with knowledge sourcesTest what the index and connectors expose; review sources.
ActDeclarative with actions · or AgenticApproval before action; log every action; least-privilege connectors.
DelegateAgenticFull agent route: identity, tool permissions, approval points, behavioural tests.
PersistAgentic, scheduled or proactiveSpecialist assurance: defined oversight, monitoring and a kill switch.
What to assure: Generative

Output quality, grounding sources, data entered, transparency labels.

T3 · T4 · T6
What to assure: Declarative

The declared configuration: instructions, knowledge sources, actions, owner, version and publishing scope. Reviewed like code.

T7 · A2 prompt and skills record
What to assure: Agentic

The whole human-agent system: identity, tools, approval points, escalation, logs, cost caps and the kill switch.

T7 · T8 · T10 · agent register

Capability and risk classification

1. The most it will do in this use case
2. What raises the stakes
3. Deterministic alternative test

Assurance route

Choose a capability class

Start with the most the tool will do in this use case, not the most it could do.

Assessment questions

0 of 60 answered · 0 concerns · 0 not known

  • Feeds: T1 Opportunity Brief · T2 Duplication Check
  • What need does the tool meet?
  • Is an approved capability already available?
  • Is AI necessary?
  • Who benefits and who may be adversely affected?
  • What evidence supports expected value?
  • Can the outcome be evaluated?

Answers stay in this browser. “Concern” and “Not known” items carry into the decision record as conditions or open questions.

Decision record outcomes

Approve

Proceed within the documented scope and controls.

Approve with conditions

Proceed once named conditions are met; conditions are tracked.

Pilot

Controlled authority for a limited population, time and data set.

Restrict

Narrow the users, data or actions permitted.

Extend

Extend a pilot to gather the missing evidence.

Stop

Do not proceed. Record the reason and lessons.

Back to the contents · Suggest a correction (opens GitHub)

Part 2 · Decide and prioritiseChapter 2.5 · 10 of 76

Criticality & assurance gates

Usage is not the same as criticality

Evaluate across ten dimensions to identify the agents that need stronger governance, assurance and operational support. Score each 1 (very low) to 5 (very high); the total sets the criticality tier and the action.

1. Usage & adoption

How many people use it and how often?

2. Audience & reach

Who relies on it? Internal teams, partners, other organisations, or the public?

3. Business impact

How important is it to achieving business outcomes or service delivery?

4. Decision impact

Does it influence, recommend or make decisions that have significant consequences?

5. Data & information

What is the sensitivity, value and regulatory significance of the data it uses or produces?

6. Autonomy level

How independently can the agent act? Human-in-the-loop versus autonomous.

7. Irreversibility

How easy is it to reverse an error or unintended action?

8. Integration & dependencies

How integrated is it with other systems, processes or agents?

9. Risk & compliance

What are the legal, regulatory, security and reputational risks?

10. Public consequence

Could failure cause harm, loss of trust or impact citizens?

CRITICALITY SCORE

— —

0 of 10 dimensions scored

Score every dimension for a criticality tier.
From score to action

12–20 LowMonitor, light governance, owner accountable

21–30

ModerateStandard governance, regular review and quality checks

31–40

HighEnhanced assurance, monitoring, human oversight required

41–50

Very highEnterprise governance, formal support, continuity and audit

Scoring guide
  • 1 Very low — Minimal impact or exposure
  • 2 Low — Limited impact or exposure
  • 3 Moderate — Noticeable impact or exposure
  • 4 High — Significant impact or exposure
  • 5 Very high — Critical impact or exposure
Example agents across tiers

LowPersonal research assistant used by creator only

ModerateTeam policy search agent used by a department

HighCase triage agent influencing case allocation and actions

Very highCitizen service agent that makes determinations or commitments

Key questions to ask
  • Who owns it and is responsible for its use?
  • What happens if it gives the wrong answer?
  • Where does it get its information from?
  • How is it tested, monitored and reviewed?
  • Does it support a critical business process?
  • Is it shared outside its original team or creator?
  • Could it become business or citizen critical?
Important reminders
High usage does not equal high criticality
Low usage can still be business critical
Reassess regularly as agents, usage and context change
When in doubt, apply the higher level of control

Right controls at the right level of risk and impact

Assurance increases with audience, autonomy, information sensitivity, criticality and consequence. Pick a tier to see the gate required before use.

AudienceWho is affected? InformationWhat data? AutonomyWhat can it do? CriticalityHow essential? ConsequenceWhat is the impact?

Tier 2 · Departmental

Departmental capability · more users, more data

Audience Department / multiple teams

Autonomy Acts with human approval

Information sensitivity Moderate

Criticality Medium

ASSURANCE GATE — REQUIRED BEFORE USE Formal assessment (risk & data review)

Key controls
  • Data source approval
  • Access controls
  • Human-in-the-loop
  • Quality checks

Oversight & accountabilityBusiness owner + data/info owner

Evidence & loggingAudit log · decisions recorded

Review frequencyMonthly

Notes on using the tiers
  • Use the lowest tier that achieves the required outcome.
  • Reassess tier when usage, data, autonomy or audience changes.
  • When in doubt, escalate the risk and apply higher-tier controls.
Assurance principles
Proportionate to risk
Risk-based and evidence-led
Human oversight at the right level
Transparent and explainable
Continuously improved

Assurance by design, proportionate by risk, continuous by evidence

A holistic, cross-disciplinary approach ensuring AI systems and agents are safe, secure, reliable, fair — and deliver public value.

Assurance domains

Security Privacy & data protection Data & information governance Ethics & fairness Transparency & explainability Human oversight & accountability Reliability & safety Evaluation & testing Auditability & evidence Risk management Legal & compliance Accessibility & inclusion

Applied across the lifecycle

1. Plan & designDefine purpose, risks, requirements and success measures

2. Build & deployDevelop, test, assure and prepare for secure deployment

3. Operate & monitorMonitor performance, risks, usage and emerging issues

4. Review & improveLearn from evidence, update controls and improve continuously

Key assurance questions
  • What are we trying to achieve and for whom?
  • What could go wrong and who could be harmed?
  • Is the data appropriate, accurate and lawful?
  • How will decisions be explained and challenged?
  • Can we evidence compliance and responsible use?
  • How will we monitor, learn and improve?
Evidence & assurance artefacts
  • Risk assessments
  • Data & DPIA records
  • Model cards & evaluations
  • Test results & red-team reports
  • Decision logs & audit trails
  • Policies, standards & guidance
  • Change logs & reviews

The human oversight and accountability chain

Agents must operate within clearly defined accountability and oversight boundaries, with named owners and transparent decision trails. Clear accountability at every step; human judgement at the right points.

1
Source owner
  • Owns the authoritative information and data made available.
  • Ensure accuracy and currency
  • Define permitted use
  • Manage access and permissions
2
Agent owner
  • Owns the agent's purpose, behaviour and operation.
  • Define purpose and scope
  • Configure behaviour and guardrails
  • Monitor performance and usage
3
Reviewer
  • Reviews AI outputs, recommendations and actions.
  • Validate accuracy and relevance
  • Check for bias, risk and compliance
  • Decide to approve, edit or escalate
4
Decision-maker
  • Accountable for the decision and its consequences.
  • Make or authorise the decision
  • Consider impact on people and services
  • Record the rationale
5
Service owner
  • Accountable for the service, outcomes and performance.
  • Ensure outcomes are achieved
  • Manage risks, incidents and issues
  • Provide assurance and improvements
6
Audit & redress
  • Provides independent assurance and holds the organisation to account.
  • Audit and review activity and controls
  • Ensure records, transparency and compliance
  • Enable challenge and redress
Human oversight points

1. Source approvalInformation is approved for use by the agent.

2. Configuration reviewAgent design and guardrails are reviewed and authorised.

3. Output reviewOutputs are checked for quality, risk and compliance.

4. Decision authorisationFinal decision is made or authorised by a person.

5. Outcome monitoringOutcomes and performance are monitored and reviewed.

6. Independent assuranceIndependent review, audit and redress are available.

Benefits

Clear accountability Better decisions Reduced risk Greater trust Stronger public value

Back to the contents · Suggest a correction (opens GitHub)

Part 2 · Decide and prioritiseChapter 2.6 · 11 of 76

Maturity & assurance · illustrative frameworks

Where you are, and how much control that needs

Two illustrative frameworks from the capability report: a seven-stage maturity model for organisations, and six assurance tiers for agents. Neither scores named organisations; both are diagnostic tools to calibrate against evidence. Over-control halts adoption; under-control creates unmanaged exposure.

The seven-stage maturity model

Observable evidence at stage 2

Bounded pilots; named sponsors; temporary funding; largely self-reported benefit evidence.

Evidence to collect before claiming progression

Named owners, repeatable gates, measured outcomes — not licensed tools.

Leadership & service ownership

Named accountable owners, clear outcomes, risk appetite and decision records.

People & workforce

Role-based training, protected learning time, inclusive adoption measures, staff consultation.

Technology & agents

Register, architecture, approved platforms, integrations, autonomy level, observability.

Data & information

Authoritative sources, metadata, permissions remediation, retention, provenance.

Security & assurance

Threat model, DPIA where required, evaluation evidence, incident route, audit logs.

Commercial & supplier

Licensing clarity, usage data, audit rights, exit, substitution, support terms.

Finance & FinOps

Forecasts, showback, anomaly alerts, unit economics, value attribution.

Community learning

Reusable artefacts, supersession, contribution workflow, evidence of reuse.

Six agent assurance tiers — controls scale with autonomy

Bounded autonomy and recoverability adopted from Government of Canada agentic-AI guidance (May 2026).

1

Personal assistantDrafting, summarisation and retrieval for one accountable user

Acceptable-use policy, permissions hygiene, human verification, retention and transparency guidance.

2

Personal or team agentBounded knowledge and repeatable team tasks

Registered owner, approved sources, test prompts, sharing gate, usage telemetry.

3

Departmental reusable agentMultiple teams or a repeatable departmental process

Business, technical, information, risk and support owners; service level; change control; retirement plan.

4

Operational agentActs on records, workflows or line-of-business systems

Distinct identity, least privilege, action logging, rollback, incident response, continuous monitoring.

5

Citizen-impacting agentInfluences access, rights, safety, services or public communication

Impact assessment, legal basis, accessibility and equality assessment, transparency, redress, independent assurance.

6

Autonomous / multi-agentDelegated goals, tools, sub-agents or long-running actions

Sandboxing, constrained authority, short-lived credentials, kill switch, behavioural monitoring, dependency mapping.

Tier finder — two questions set the tier at intake

What may this agent do without asking, and what can be undone when it goes wrong?

Back to the contents · Suggest a correction (opens GitHub)

Part 3 of 12 · 8 chapters

Design and build

Part 3 · Design and buildChapter 3.1 · 12 of 76

Responsible AI · implement AI

Twelve stages, twelve gates, one evidence chain

Each stage answers one question, produces one artefact and passes one gate. Select a stage to filter the templates. The Agent journey chapter follows a single agent through the same path.

Explore Assess and design Deploy and enable Operate and evolve

Key gates

  1. Validated need

    Stage 1

  2. Reuse decision

    Stage 2

  3. Assessment approval

    Stage 4

  4. Pilot authority

    Stage 6

  5. Deployment approval

    Stage 9

  6. Continued authority

    Stage 10

  7. Closure confirmed

    Stage 12

Where each phase lives, what it produces, who does it

Portal sectionKey outputsPrimary roles
Start hereOpportunity brief · initial questions · support and contactAll colleagues
Find an AI toolCatalogue search · tool comparison · reuse decisionUsers · IT and digital
Assess a use caseRisk classification · data and privacy assessments · supplier and model review · approval recordProject team · risk, security, privacy · data and information
Implement AIDesign record · pilot plan · evaluation report · deployment checklistProduct owner · technical team · users and change leads
Operate and monitorUsage, cost and outcome data · incident and risk records · change-impact assessments · continuous improvementService owner · operational support · governance and assurance
Learn and reuseLessons learned · updated guidance · patterns and templates · retirement recordAll teams · community of practice · centre of excellence

Templates and artefacts · 12 core, 7 added

T1 · Core · stage 1

AI Opportunity Brief
  • Reference, title, owner and team
  • Problem evidence and affected users
  • Current process and non-AI options
  • Desired outcome and success measures
  • Inputs, outputs, decisions and systems
  • Accessibility need, urgency and constraints

T2 · Core · stage 2

Existing-Capability and Duplication Check
  • Required capability
  • Catalogue and pattern searches
  • Existing tools, agents, prompts and skills reviewed
  • Gap and configuration assessment
  • Reuse, extend, build or buy rationale
  • Decision and consulted owners

T3 · Core · stage 3

Capability and Risk Classification
  • Ask, Create, Connect, Act, Delegate, Persist
  • Information sensitivity and personal data
  • System access and autonomy
  • Consequential decisions and vulnerable users
  • Scale, reversibility and public visibility
  • Risk tier and assurance route

T4 · Core · stage 4

Information and Data Readiness Assessment
  • Sources, owners and classification
  • Personal data and DPIA needs
  • Permissions and business need
  • Metadata, filenames, titles and indexing
  • Quality, authority, retention and deletion
  • Traceability, aggregation and training use

T5 · Core · stage 4

Supplier, Model and Deployment Assessment
  • Supplier, edition and deployment route
  • Models, provider and automatic selection
  • Subprocessors, locations and retention
  • Contract, policy and admin controls
  • Logging, change notice and deprecation
  • Exit plan and unresolved uncertainties

T6 · Core · stage 4

Responsible AI and Human Oversight Assessment
  • Task and decisions affected
  • Reviewer, competence and evidence access
  • Uncertainty, challenge and override
  • Automation bias and prohibited uses
  • Failure, refusal and conflict handling
  • Redress and oversight effectiveness

T7 · Core · stage 5

Agent Design Record
  • Purpose, owner, identifier and status
  • Model, instructions, skills and knowledge
  • Identity, permissions, tools and credentials
  • Triggers, schedules, actions and other agents
  • Approval points, escalation and stopping
  • Tests, logging, monitoring, kill and review

T8 · Core · stage 6

Pilot Plan
  • Hypothesis, population, duration and environment
  • Permitted data and tasks, and exclusions
  • Support and training
  • Quality, accessibility, cost, trust and risk measures
  • Comparison and baseline design
  • Stop criteria and evidence ownership

T9 · Core · stage 7

Evaluation and Benefits Record
  • Outcomes, quality and reliability
  • Accessibility, wellbeing and user outcome
  • Time, review and rework
  • Cost, adoption and sustainability
  • Unintended effects and limitations
  • Evidence status and recommendation

T10 · Core · stage 9

Deployment Readiness Checklist
  • Catalogue and ownership
  • Assessments and approvals
  • Permissions and indexing tests
  • Oversight and accessibility
  • Guidance, support and monitoring
  • Incident, kill, retention, review and retirement

T11 · Core · stage 11

Change-Impact Assessment
  • Trigger: model, supplier, data, permission, action, interface, cost or policy
  • Impact on risks and controls
  • Evidence and documents to update
  • No, partial or full reassessment
  • Continue, restrict, suspend or retire decision

T12 · Core · stage 12

Retirement Record
  • Reason and replacement
  • User communication
  • Disable agents and actions, and revoke access
  • Disconnect integrations and dispose of data
  • Update records and supplier access
  • Residual risks, lessons and closure approval

A1 · Added · stage 9, 10

Adoption and capability plan
  • Senior sponsor and local champions
  • Protected learning time
  • Local communications
  • Accessible training and reinforcement
  • Support

A2 · Added · stage 5, 10

Prompt and skills product record
  • Owner, version and purpose
  • Approved users
  • Test set and security checks
  • Cost profile
  • Retirement

A3 · Added · stage 5, 9

Support and service record
  • Support route and service dependencies
  • Escalation
  • Known limitations
  • Incident response
  • Service-level position

A4 · Added · stage 6, 7

Comparative task-test template
  • Baseline tool versus proposed capability
  • Task, time and cost
  • Review effort and output quality
  • Failure modes
  • Accessibility

A5 · Added · stage 4, 7

Ways-of-working impact assessment
  • Delegation and role development
  • Workload and meeting volume
  • Professional review
  • Change-management effects

A6 · Added · stage 5, 9

Guest and cross-tenant access plan
  • Resource location
  • Permissions
  • Alternate access route
  • Accessible distribution

A7 · Added · stage 11

Release and guidance change record
  • Product terminology and screenshots
  • Training assets
  • Affected audiences
  • Communication date

Back to the contents · Suggest a correction (opens GitHub)

Part 3 · Design and buildChapter 3.2 · 13 of 76

Design and build · capability classes

Generative, agentic and declarative

Three ways of using the same underlying models, with different failure modes and therefore different controls. Most real deployments combine all three, which is exactly why the class needs naming before the controls are chosen.

Generative

Produces content, analysis or options on request. Nothing changes until a person acts on the output.

You specify: A prompt, grounding sources and an output format.

Drafting, summarising, translating Analysis and option generation Search and question answering Code and test generation

Primary failure Plausible but wrong output accepted without review. Tier tendency Usually low to moderate: consequence sits with the person who acts on the output.

Controls that carry the weight

  • Accuracy and grounding evaluation
  • Citation and provenance
  • Disclosure that AI was used
  • Human review before use

Agentic

Takes actions across tools and systems under delegated authority, deciding its own path to a goal.

You specify: A goal, tools, permissions, guardrails and approval thresholds.

Triage and routing Multi-step workflow execution System updates within limits Orchestration across services

Primary failure Correct-looking action taken outside intended scope, or an irreversible action taken without approval. Tier tendency Moderate to high: the agent itself changes the world, so authority and reversibility dominate.

Controls that carry the weight

  • Capability separated from authority
  • Least-privilege permissions and recertification
  • Approval gates for consequential actions
  • Full action logging and rollback

Declarative

You state the outcome, rules and boundaries; the platform resolves how they are met and keeps behaviour inside the declaration.

You specify: Instructions, knowledge sources, permitted actions and constraints — as configuration, not code.

Configured assistants defined by a manifest Policy and rules expressed as configuration Low-code flows with declared triggers Templated agents published to a team

Primary failure The declaration drifts from intent, or nobody owns the specification that is now running in production. Tier tendency Varies with what is declared: the specification is the control surface, so review it as you would code.

Controls that carry the weight

  • Version control over the declaration
  • Named owner for the specification
  • Change assessment on every edit
  • Regression tests against the declared outcome

Side by side

Generative Agentic Declarative

What you supply Prompt and sources Goal, tools and authority Instructions, sources and constraints

Who chooses the path The person The agent, within guardrails The platform, within the declaration

What changes Nothing until a person acts Systems and records directly Whatever the declaration permits

Primary failure Wrong content believed Wrong action taken Wrong specification running

Assurance emphasis Output quality and provenance Permissions, approval and rollback Change control over the declaration

Register fields that matter most Sources, evaluation, review point Permitted actions, authority level, escalation Version, owner, constraints, publication scope

Which class is this?

Three questions. The answer sets the controls that matter most — not whether the work is allowed.

Does it change anything outside the conversation?
Who decides the steps?
Can its effects be undone without a person?

Most deployments are all three at once

Declarative shell

The declaration sets purpose, sources, permitted actions and limits — and is the artefact under change control.

Generative core

Inside those limits, the model drafts, summarises and reasons; evaluation and provenance apply here.

Agentic actions

Only declared actions run, each with its permission, approval threshold and audit trail.

Record the class in the register alongside purpose and authority. It tells a reviewer which controls to look for before they open the evidence — and it makes the difference between an agent that drafts and an agent that acts visible without reading the configuration.

Back to the contents · Suggest a correction (opens GitHub)

Part 3 · Design and buildChapter 3.3 · 14 of 76

Skill studio · modular instruction block

Build the instruction. Inject the skill.

A reusable way to brief any assistant: seven parts, a directives layer that always applies, validation that is always on, and a library loop so good prompts get kept, shared and versioned — not rediscovered.

The block — seven parts

Skill focus — pick any

System directives — always applied

Be accurate & truthful Be helpful & relevant Be safe & secure Be fair & inclusive Cite sources when possible Declare limits & assumptions

Assembled block

SYSTEM DIRECTIVES: be accurate and truthful; be helpful and relevant; be safe and secure; be fair and inclusive; cite sources when possible; declare limits and assumptions.

Validation — always on · 0 of 8 passed

Governance & safety rails

Confidentiality & privacy. respect data boundaries and permissions

Security & compliance. follow policies, standards and legal requirements

Trust & transparency. show reasoning, sources and confidence levels

Fairness & non-discrimination. avoid bias; promote equity and inclusion

Human oversight. AI supports; humans decide and take accountability

Monitoring & audit. logs, reviews, feedback and continuous improvement

Risk & harm prevention

Misinformation Privacy breach Overreliance Bias & exclusion Unsafe content Hallucination

DEIA considerations

Accessibility by design Inclusive language Cultural & regional awareness Multiple perspectives Bias detection & mitigation Cognitive load & readability

Learning & improvement loop

Use

Apply the prompt in real work

Review

Evaluate output and experience

Refine

Improve prompt and instructions

Capture

Add to library; share learnings

Evolve

Retire, update or version prompts

The reusable library holds role-based, task-based and domain prompts, scenario templates, anti-pattern examples — tested and versioned, with supersession history.

Success signals

Useful Trusted Inclusive Accessible Traceable Actionable Impactful

Principles at the core: human-centred · inclusive by design · transparent · accountable · reliable · respectful · secure · sustainable · equitable.

Back to the contents · Suggest a correction (opens GitHub)

Part 3 · Design and buildChapter 3.4 · 15 of 76

Responsible AI · agents and skills

Assure the whole human-agent system, not only the output

What starts an agent, how it plans, what it can reach, where people stay in control, and what it leaves behind. Skill studio builds the instruction block; Autonomy boundaries sets the limits.

1. Trigger

What starts the agent?

  • User request
  • Scheduled task
  • Event: new email, meeting or file
  • System trigger: ticket or alert
  • API or webhook
  • Another agent
2. Agent core

How it thinks, plans and decides

  • Instructions and goals: system prompt, skills, guardrails
  • Model: chosen models and settings
  • Knowledge and context: organisational data, memory, external sources
  • Reasoning and orchestration: plans tasks, selects tools, handles errors
3. Tools and actions

What it can reach and do

  • Search and retrieve across approved sources
  • Call APIs and applications
  • Delegate to other agents
  • Create and update content, records, messages and code
4. Human oversight

Where people stay in control

  • Approval before high-risk actions
  • Review and validation of outputs and sources
  • Challenge and feedback
  • Stop or override at any point
5. Outcome

What it delivers

  • Completed task or output
  • System update or action taken
  • Notification or communication
  • Log and evidence: what happened, when and why

Cross-cutting control rails

Identity and permissions

Who the agent is and what it can reach

Information and data controls

Classification and retention

Monitoring and logging

Activity, actions and decisions

Cost and consumption

Budgets, limits and alerts

Security and privacy

Risks, DPIA and incident response

Accessibility and inclusion

Usable for all

Sustainability

Environmental impact

Kill switch and retirement

Stop and close safely

Risks the agent route must test

Prompt and skill supply chain

A shared skill or instruction changed without review changes every agent that loads it.

Proposed

External input and prompt injection

Documents, web pages and emails can carry instructions that redirect the agent.

Proposed

Identity, delegation and shared accounts

Agents acting through shared or delegated contexts can reach more than the requester could.

Reported

Unexpected discovery

Permissions alone may not stop titles, filenames and metadata surfacing through the index.

Reported

Multi-agent orchestration

Agents calling agents need identity, approval points and automated records at each hop.

Reported

Telemetry gaps

Agent activity logs may not show enough to reconstruct what happened.

Reported

Continuous assurance

  1. 1. Test before deployment

    Behaviour, safety and access

  2. 2. Monitor in operation

    Usage, outcomes and incidents

  3. 3. Learn and improve

    Feedback and evaluation

  4. 4. Reassess on change

    Model, tools, data and permissions

  5. 5. Retire when no longer needed

    Revoke access, close data, capture lessons

Prompts and skills as organisational products

A personal prompt stays personal. Once a prompt, skill or agent instruction is shared, it needs an owner, a version, a test set, security checks, a cost profile and a retirement date, and it goes in the prompt and skills register.

  • Is the prompt or skill personal or organisational?
  • Does it need version control and approval?
  • Has it been tested for safety, consistency and cost?
  • Could deterministic automation do the job better?
  • Can it be shared or changed without review?
  • How will it be withdrawn?

Deterministic alternative test

Before building a generative skill or agent, check whether conventional automation is safer, cheaper and more consistent.

If a rule, form, flow or script can do the job reliably, use it. Generative AI earns its place where judgement over unstructured material is the actual task.

Back to the contents · Suggest a correction (opens GitHub)

Part 3 · Design and buildChapter 3.5 · 16 of 76

Design and build · connectors, MCP and integrations

Every connector is a new data path and, often, a new action

MCP servers, connectors, plugins, add-ins and APIs turn a generative tool into one that can reach, and change, other systems. Each needs an owner, a scope and a log.

MCP serverAn open-protocol endpoint that exposes tools and resources to any MCP-capable client. Tools can be discovered at runtime, so capability can change without the agent being republished.

ConnectorBrings an external system’s content into an AI tool’s index or grounding, such as Copilot connectors.

Plugin or actionA packaged custom action, such as an MCP or API plugin in a declarative agent, that lets an agent read or change data.

Add-in or appAn extension that runs inside a host application, such as an Office add-in or a chat app.

API and webhookDirect integration through REST APIs, OpenAPI descriptions or event webhooks.

Integrator or gatewayA platform that brokers many connections, applies policy and logs traffic, such as an MCP gateway or integration platform.

Platform directory

Graded claims from supplier documentation. Where a route is unresolved, confirm it with the supplier before connecting.

Microsoft 365, Copilot and Copilot Studio Verified

Declarative agents call MCP servers and REST APIs through MCP and API plugins. Copilot Studio adds MCP servers through an onboarding wizard, and servers can be registered in Agent 365.

Routes: MCP plugin · API plugin · Copilot connectors · Power Platform connectors

Controls: Tools are discovered at runtime unless pinned in the manifest. Register MCP servers in Agent 365 and review each tool’s read and write scope.

MCP and API plugins for declarative agents ↗
Connect Copilot Studio to an MCP server ↗
Declarative agents for Microsoft 365 Copilot ↗

GitHub Verified

GitHub maintains a remote MCP server for repositories, issues and pull requests. Microsoft’s walkthrough wraps it as a plugin in a declarative agent.

Routes: Remote MCP server · GitHub Apps · Actions

Controls: One OAuth app per organisation, repositories scoped, no write tools without approval, and no secrets in prompts.

Build a declarative agent plugin from the GitHub MCP server ↗
Atlassian: Jira, Confluence, Jira Service Management, Bitbucket, Loom

Corroborated

The official Rovo MCP server reached general availability on 4 February 2026. It is cloud-hosted and authenticates with OAuth 2.1 or API tokens.

Routes: Remote MCP server, v2 endpoint · Rovo agents

Controls: Admins choose which AI tools and domains may connect, destructive tool groups stay off until enabled, and every tool call is written to the audit log. Atlassian Cloud only, not Data Center.

Getting started with the Rovo MCP server ↗
atlassian-mcp-server on GitHub ↗

Anthropic Claude Reported

Claude connects to Microsoft 365 and other systems through connectors built on MCP. Its Microsoft 365 connector reads Outlook, Teams, SharePoint and OneDrive and writes to the last two.

Routes: Connectors · remote MCP servers · Skills

Controls: Set connector write scope per user, keep an admin allow-list, and record each connector in the data-source register.

No verified documentation linked yet.

Google Workspace and Gemini Unresolved

Gemini works with Workspace content and connected data. Confirm the current MCP and extension routes with Google before use.

Routes: Workspace grounding · extensions · APIs

Controls: Confirm data location, admin controls and audit logging for each extension.

No verified documentation linked yet.

Slack Unresolved

Slack can be a source or a destination for agent work. Confirm the official MCP or app route and its data residency before connecting.

Routes: App · MCP server · webhooks

Controls: Scope channels, keep private channels out without approval, and align retention.

No verified documentation linked yet.

Service and line-of-business systems Proposed

ServiceNow, Salesforce, Dynamics 365 and case-management systems publish connectors, APIs or MCP servers. Each is a new data path and a new action surface.

Routes: Connector · API plugin · MCP server · integration platform

Controls: Treat any write tool as Act class, require approval before change, and log to the incident trail.

No verified documentation linked yet.

Model Context Protocol, the standard Corroborated

An open standard that lets an AI client call tools and read resources on an outside system through one interface.

Routes: Specification · registry · gateways

Controls: Prefer official servers, pin tools where possible, review tool descriptions for injected instructions and route calls through a gateway that logs them.

modelcontextprotocol.io ↗

Nine controls for every connector, plugin and MCP server

  1. Register every connector, plugin and MCP server in the tool and data-source registers, with a named owner.
  2. Record read and write scope separately. Any write tool makes the use case at least Act class.
  3. Prefer official, supplier-maintained servers over community ones, and record which is in use.
  4. Pin tools or versions where the platform allows. Dynamic discovery means capability can change without republishing.
  5. Authenticate with OAuth and least privilege. No shared service accounts for user-level data.
  6. Log every tool call to an audit trail the service owner can read.
  7. Review tool descriptions and returned content for prompt injection.
  8. Check where each connected system stores and processes data, and whether it sits inside your data boundary.
  9. Keep a revoke route: users can withdraw authorisations and admins can disable a connector at once.

A connector can change the capability class

A generative chat tool with a read connector is Connect class. Add a write tool and it is Act class; let it run on a schedule and it is Persist. Reclassify whenever a connector or tool is added, and record it in the change register.

Back to the contents · Suggest a correction (opens GitHub)

Part 3 · Design and buildChapter 3.6 · 17 of 76

Architecture & quality

Principles with tests, gates with teeth

A three-tier canon governs everything that influences how something is built — prototypes explicitly included, because most principle breaches enter the estate through prototypes that were never intended to survive, and then did.

Tier 1 — mandated, cited not restated

T1-01 Secure by Design T1-02 CAF & GovAssure T1-03 Technology Code of Practice T1-04 GDS Service Standard T1-05 AI Playbook T1-06 Data protection & public records T1-07 WCAG 2.2 AA

Tier 1 conformance is binary: the principle applies and is met, or a recorded exception exists. No informal partial conformance.

Tier 2 — the core seven, each with a test

AP-01
Federated, not centralised

Test: Can each element be traced to a named authoritative owner outside the central platform?Looks compliant, is not: "We will federate later, once everything is in one place."

AP-02
Discovery separated from access

Test: Can a user discover an asset exists without thereby gaining access to it?Looks compliant, is not: Endpoint or credential details published as discovery metadata for convenience.

AP-03
Human-in-the-loop by design

Test: Can any automated component write to the record without recorded human approval?Looks compliant, is not: "The user can always review it" — optional review presented as a control.

AP-04
Traceability & provenance mandatory

Test: Take any output at random: can you reconstruct inputs, owner, decision and evidence without asking a person?Looks compliant, is not: Evidence linked at closure rather than at capture.

AP-05
Reuse before build

Test: Is there a recorded reuse assessment that predates the build decision?Looks compliant, is not: Reuse assessment performed after the build decision, to justify it.

AP-06
Accessible & inclusive by design

Test: Is accessibility evidence produced before release? Does every status carry a non-colour cue?Looks compliant, is not: "We will run an accessibility audit before go-live" as the only accessibility activity.

AP-07
Guardrails enforced, not documented

Test: Attempt to breach the control through the normal interface. Does it succeed?Looks compliant, is not: A mandatory gate with no exception path — guaranteeing informal workarounds.

Prototype promotion gate. A prototype may not be promoted toward live use until Tier 1 and Tier 2 conformance is assessed and recorded. Record intended lifespan and disposal point at creation; record deferred principles; re-assess them at the gate. Human-in-the-loop applies in full even at prototype stage — autonomy introduced early is rarely removed later. Promotion is a decision, not a drift.

The assurance-gated agentic pattern

From the AI Engineering Lab's reference implementation: the transferable innovation is not any one use case — it is the operating model around it. Specialist agents routed by an orchestrator; a challenger agent that reviews outputs; a compliance agent; assurance gates executed in the workflow; immutable audit trails; human approval before anything writes to the record. The same pattern serves capability management, catalogue operations, workforce planning and architecture governance.

Specialist agents Orchestrator routing Challenger agent Compliance agent Assurance gates in-workflow Immutable audit trail Human approval to write Escalation paths

The quality gate — score prototypes like you mean it

A gate, not a form: run at the end of every iteration. Tier 0 is the brake — the only tier that can argue for less; a veto is the instrument doing its job. Scores are capped by evidence strength, unscoreable items go on an open register published with the score, and demo drift — a level-3 prototype someone useful quietly relies on — is the most common governance failure in this class of work.

Quick index — ten dimensions, 1–5

Security
Privacy & data protection
Accessibility
Usability
Performance & resilience
Maintainability
Observability
Data quality & provenance
Trust & transparency
Sustainability & handover
—
Sophistication index — Σ scores ÷ 50 × 100

Score the dimensions for a reading

The ladder: Tier 0 proportionality gate → Tiers 1–6 architecture compass → Tier 7 self-* autonomic → Tier 8 human & institutional sustainability. Evidence-rung caps; accessibility floors at human testing, not automated checks; an empty removal ledger three iterations running is itself a Tier 0 veto. Dimensions here are representative — attach the full template for real assessments.

Back to the contents · Suggest a correction (opens GitHub)

Part 3 · Design and buildChapter 3.7 · 18 of 76

Knowledge foundation

From weekly call to institutional memory

Transforming conversations into trusted, discoverable knowledge that drives action and public value. Capture once, reuse everywhere — the seven-step pipeline and the record it produces.

1. Capture

Capture the moment
  • Record meeting
  • Capture chat
  • Collect shared files
  • Note live questions
  • Identify demos
Raw recording, chat and artefacts
2. Transcribe
Make it searchable
  • Generate transcript
  • Speaker identification
  • Time-stamped content
  • Redact sensitive information
Transcript with timestamps
3. Summarise
Extract the essence
  • Executive summary
  • Key points
  • Decisions
  • Unresolved questions
  • Action items
Structured summary and outcomes
4. Curate
Add context and quality
  • Add metadata and taxonomy
  • Link to related topics
  • Classify sensitivity
  • Verify facts and sources
  • Assign owners
Curated, classified session record
5. Connect
Link and enrich
  • Link to use cases, agents and data assets
  • Classify sensitivity
  • Verify facts and sources
  • Assign owners
Connected knowledge with relationships
6. Publish
Make it discoverable
  • Publish to knowledge hub
  • Update indexes
  • Set permissions
  • Notify communities
  • Version control
Discoverable knowledge asset
7. Reuse & learn
Drive action and value
  • Apply in projects and services
  • Share improvements
  • Capture feedback and results
  • Update and improve
Impact, feedback and new learning
What we capture

Discussions and insights Questions and challenges Decisions and agreements Actions and owners Shared links and resources Demos and use cases Risks and issues Ideas and opportunities Lessons learned

Enablers of institutional memory
  • Clear governance and ownership
  • Consistent metadata and taxonomy
  • Security, privacy and access control
  • Quality assurance and validation
  • Retention and lifecycle management
  • Community curation and care
  • Usage analytics and feedback
  • Integrated tools and platforms
The benefits

Easier to findSearchable, organised knowledge at your fingertips

Avoid duplicationLearn from others' experience and mistakes

Stronger decisionsBetter informed choices backed by evidence

Faster deliveryReuse patterns, guidance and assets to go faster

Assured and trustedQuality, governance and provenance you can rely on

Greater public valueBetter services, outcomes and trust for citizens

The session evidence pack

Twelve components that turn a conversation into a reusable organisational asset. Session record · community owned · governed and secure · discoverable and reusable · continuously improved.

1. Session details Title & description · Date, time, duration · Community / forum · Participants & roles · Host / facilitators2. Recording & transcript Recording link · Transcript / captions · Time-stamped version · Accessibility info3. Summary & key points Executive summary · Key discussion points · Important context · What was agreed4. Questions & answers Questions raised · Answers provided · Outstanding questions · Follow-up needed
5. Demonstrations & artefacts Demos & walkthroughs · Screens / diagrams · Code / configs · Resources shared6. Decisions & agreements Decisions taken · Agreements reached · Scope confirmations · Changes approved7. Actions & owners Action items · Owners · Due dates · Status / progress8. Linked resources & references Related publications · Previous sessions · External references
9. Classification & sensitivity Information classification · Sensitivity level · Handling instructions · Access permissions10. Evidence & insights Evidence referenced · Impact / value assessed · Metrics mentioned · Outcomes expected11. Lessons & insights Key takeaways · Update history · Supersedes / superseded · Recommendations12. Review & lifecycle Review cycle · Update history · Supersedes / superseded · Retention
How it flows
  • 1. Capture — Capture during or after the session
  • 2. Structure — Organise using the standard evidence pack template
  • 3. Classify — Apply classification, tags and sensitivity controls
  • 4. Publish — Store in governed repositories and make discoverable
  • 5. Link — Connect to related assets, agents, use cases, and more
  • 6. Reuse — Teams find, reuse and build on the knowledge
  • 7. Improve — Feedback and updates improve content continuously
Evidence status on every claim

VerifiedIndependently confirmed

Reported / emergingReported by a source, not yet validated

In progress / unvalidatedBeing worked through

Contradictory / needs reviewSources disagree

Outdated / supersededReplaced by newer content

The knowledge lifecycle

Good knowledge is captured once, curated well, reused often and improved continuously — and its status is always visible.

Create & capture 1. Capture Capture knowledge, evidence and experiences from sessions, projects or operations.

Create & capture 2. Classify & tag Add metadata, taxonomy, topics, sensitivity and links to related assets.

Curate & govern 3. Curate & quality Check accuracy, completeness, relevance and quality. Remove duplicates.

Curate & govern 4. Validate & approve Subject matter and governance review. Approve for use and assign lifecycle status.

Share & use 5. Publish & discover Publish to trusted repositories and make discoverable through search, portals and APIs.

Share & use 6. Use & reuse Apply knowledge in work, decisions and services. Share feedback and outcomes.

Share & use 7. Review & update Review on a regular cycle or when triggered by change. Update content, links, owners and status.

Share & use 8. Sunset, supersede or archive Retire outdated content, mark as superseded or archive with retention and audit trail.

Evidence status through the lifecycle

Draft Newly captured. Not yet reviewed.

Reviewed Peer or owner reviewed. Accuracy and relevance checked.

Validated Approved by subject matter and governance for use.

Published Officially published and visible to intended audience.

Superseded Outdated or replaced by newer content. Retained for reference.

Archived No longer active. Retained in accordance with policy.

One discovery layer, many trusted sources

Information is scattered across teams, tools and tenancies, which causes duplication, inconsistency and poor discoverability. Federate rather than migrate: connect the sources, add a metadata and governance layer, and put one discovery layer on top.

Today — fragmented and hard to find

Chats & calls SharePoint sites Personal OneDrives Email & attachments Departmental file shares GitHub repositories Slack / other channels Knowledge hub Spreadsheets & lists Recordings & transcripts Documents & guidance External links & references

Target — federated knowledge service

Unified discovery layerSingle search & navigation · Smart filters & facets · Recommendations & related content · Personalised views

Metadata & governance layerCommon taxonomy · Metadata standards · Classification & sensitivity · Ownership & accountability · Lifecycle management

Federated source connectorsSource systems remain in place — connect, do not migrate

Tomorrow — connected and valuable

Find fasterOne place to search across the estate with better results

Trust & assuranceClear ownership, quality signals, classification and provenance

Reuse & collaborateEasily share, reference and reuse patterns, guidance and code

Learn & improveInsights and lessons flow back to improve practice continuously

Deliver more valueBetter decisions, less duplication, faster delivery and stronger outcomes

The agent information supply chain

High-quality information in, trusted retrieval and reasoning, accurate answers or actions, human review, better sources. Every link has a quality gate.

1. Authoritative sources
  • Trusted, approved information assets
  • Policies & procedures
  • Knowledge bases
  • Case law & guidance
  • Data & reference sets
  • Forms & templates
  • Expert directories
2. Metadata & discovery
  • Make content findable and understandable
  • Metadata & tagging
  • Taxonomy & keywords
  • Semantic models
  • Lineage & provenance
  • Classification & sensitivity
  • Permissions & access
3. Retrieval
  • Find the right information in the right context
  • Hybrid search (keyword + semantic)
  • Filters (date, owner, source, role)
  • Rank by relevance & authority
  • Permission check
4. Agent reasoning & processing
  • Synthesise and reason using retrieved content
  • Understand intent
  • Ground in source content
  • Reason & generate
  • Cite sources
  • Guardrails & policies
5. Response / action proposal
  • Answer with citations, drafts or recommended actions
  • Answer with citations
  • Draft content
  • Suggest next steps
  • Propose system actions
6. Human review & oversight
  • Humans verify, approve and take responsibility
  • Check accuracy
  • Validate sources
  • Confirm applicability
  • Decision logged
  • Approve / reject
7. Feedback & capture
  • Capture feedback, use and outcomes
  • Was it useful?
  • Was it correct?
  • What was missing?
  • New knowledge identified
  • Add & monitor
8. Source improvement & governance
  • Improve sources, policies and models
  • Update content
  • Fix gaps
  • Improve metadata
  • Retire outdated
  • Review & approve
Key metrics for the chain

Answer accuracy rate Citation coverage User satisfaction score Time to answer Re-use rate of content Feedback captured Source improvements

Where should the information live?

Work down the questions in order. The first "yes" is your answer — the goal is the right information, to the right people, at the right time, securely and efficiently.

1. Is the information already in an authoritative source (for example an existing SharePoint site, policy library, DMS)?

Yes → Use existing authoritative location — keep content in place, ensure agent uses existing permissions, maintain single source of truth, lower duplication and maintenance risk.

2. Can the agent be granted access via existing permissions?

Yes → Grant access via existing permissions — respect least privilege, use groups and roles already in place, review access periodically.

3. Can permissions be adjusted for a new site (or team) without creating excessive oversharing or security risk?

Yes → Adjust permissions or create a secure location — create dedicated site or library if needed, configure granular permissions, review with the information owner.

4. Is copying or moving content necessary for the agent to function effectively?

Yes → Copy or move content with strong controls — risk of duplication and divergence, increased maintenance and update process, maintain links back to source of truth, confirm retention and records handling.

If every answer is no — reconsider the approach
  • Use metadata, views, filters or connectors
  • Use automated search over authoritative sources
  • Build a knowledge layer or index
  • Avoid unnecessary duplication
Key considerations
  • Audience — Who will use the agent and what access do they need?
  • Permissions — Does the agent respect existing SharePoint permissions?

Information governance — Is the content authoritative, current and controlled?

Duplication risk — Will copying create multiple versions or conflicting answers?

Maintenance — Can the team keep copies current and in sync with the source?

Licensing & cost — Do users have the right licences? Any connector or storage costs?

Records & retention — Where is the record of business? How long must it be kept?

Security & risk — Does the agent minimise risk of overexposure or data leakage?

A knowledge graph and a shared prompt layer

Connecting people, communities, data, decisions and capabilities to power trusted AI — and a common language layer that improves responses, saves time and builds organisational memory.

What the graph connects
Questions it answers
  • Who owns this AI agent and what use cases is it used for?
  • What data sources ground the answers provided by this agent?
  • Which policies and controls apply to this use case?
  • What outcomes are we expecting and how are we measuring them?
  • What other capabilities or agents could be reused?
Why a knowledge graph

Connects what sits apartLinks people, data, agents and outcomes

Improves discovery and reuseFind what already exists before building

Provides context for AIGrounded, related, permissioned context

Supports governance, assurance and valueOwnership, lineage and measurable benefit

1. Organisational prompt library Curated collection of high-quality prompts · Categorised by task, role and domain · Example prompts and use guidance

2. Prompt lab & testing Experiment and refine prompts safely · Compare results and measure quality · Promote best prompts to the library

3. Business glossary & acronym dictionary Common definitions and terminology · Acronyms, abbreviations and plain-government terms · Versioned and community-maintained

4. Domain & policy knowledge packs Topical guidance, policies, frameworks · Approved content summaries for AI use · Content packs for specific domains

5. Technology-agnostic prompt patterns Reusable patterns that work across tools and models · Task-based templates and structures · Adapts to different AI platforms and models

How the layer works

1. Create & contribute Users contribute prompts, definitions and knowledge

2. Curate & review Experts review for quality, accuracy and safety

3. Publish & tag Approved items are tagged and published to the layer

4. Use & apply AI tools and users discover and apply in their work

5. Feedback & improve Results and feedback feed back to improve content

Back to the contents · Suggest a correction (opens GitHub)

Part 3 · Design and buildChapter 3.8 · 19 of 76

Responsible autonomy boundaries

The right level of autonomy for the right task at the right risk

Autonomy is a privilege, not a default. Human oversight increases with risk, impact and consequence — and the human role is named at every level, not implied.

Human controlPeople remain accountable

Proportionate by riskMatch autonomy to impact

Transparent & explainableMake actions understandable

Safe by design & defaultGuardrails built in

Continuously monitoredDetect, learn and improve

LOW AUTONOMY HIGH AUTONOMY

Level 2 of 6

Recommend

Suggest options

Agent recommends options or actions for human consideration.

Examples
  • Recommend next action
  • Suggest content or draft
  • Flag risks or anomalies
The human role
Human decision

User reviews and decides

Factors that increase autonomy
  • Low risk to people
  • Reversible outcomes
  • High data quality
  • Clear objectives
  • Strong monitoring
Factors that reduce autonomy
  • High impact or risk
  • Irreversible outcomes
  • Sensitive data
  • Uncertain environments
  • Weak monitoring
Decision factors
Risk & impact
Legal / regulatory
Data sensitivity
Complexity
Frequency
Public trust
Guardrails always on, and the override set
0 of 7 guardrails and 0 of 4 override controls in place. Guardrails are always on — autonomy is a privilege, not a default.

GUARDRAILS (ALWAYS ON) ESCALATION & OVERRIDE

Accountability
Named owners for agents
Clear responsibility for decisions
Traceable actions and outcomes
Key reminder
Increase autonomy only when the value outweighs the risk, controls are effective and humans can intervene.

Back to the contents · Suggest a correction (opens GitHub)

Part 4 of 12 · 8 chapters

Assure and govern

Part 4 · Assure and governChapter 4.1 · 20 of 76

Responsible AI · govern and assure

One control plane for any tool, model or agent

Consistent principles, proportionate assurance and reusable components. Evidence and transparency records sit with the controls because they are how each control is shown to work.

Principles and risk

Capability-based classification sets the assurance depth.

Information and data

Readiness, metadata, indexing, provenance and retention.

Security and privacy

Threat model, DPIA, transfers, subprocessors and DLP.

Human oversight

Reviewer competence, evidence access, override and stop.

Accessibility and inclusion

Accessibility of AI, AI for accessibility, alternatives and adjustments.

Cost and FinOps

User-visible cost, caps, alerts, budget owners and value criteria.

Sustainability

Supplier evidence, usage indicators and proportionate use.

Transparency and records

AI involvement statements, labelling and audit evidence.

Monitoring and assurance

Control effectiveness, incidents and reviews.

Change and lifecycle

Change watchlist, reassessment triggers and retirement.

Who owns what

RoleAccountable for
Senior responsible ownerRisk acceptance, sponsorship and the decision to proceed or stop.
Service ownerDay-to-day operation, monitoring, support and retirement.
Technical ownerConfiguration, identity, permissions, release and change.
Commercial ownerContract, licences, consumption, supplier commitments and exit.
Assurance ownerAssessments, evidence, exceptions and reviews.
Budget ownerCaps, alerts and spend against value evidence.
Agent, prompt or skill ownerVersion, tests, approved users and withdrawal.

Exceptions and the shadow AI route

Unmanaged external use falls when asking for an approved alternative is quicker than working around the rules.

  1. Search your organisation's catalogue for an approved alternative.
  2. If none fits, request one through a use case with the reason it is needed.
  3. Get an interim decision within an agreed time, with conditions.
  4. Record the exception in the decision register with an expiry date.
  5. Review at expiry: approve, restrict or close.

Communication authenticity

Keep clear labels on AI-sent messages and disclose material AI involvement. Removing a label is prohibited where transparency is required.

Evidence and transparency records

AI involvement statements

What AI did in producing a document, decision or message.

Source records

The sources an output drew on, so a reader can check them.

Human verification

Who checked the output, against what, and when.

Decision-support records

How AI informed a decision and who made it.

Code records

Provenance and review of AI-generated code.

Agent action logs

What an agent did, when and why.

Audit evidence

Configuration, approvals and test results for assurance.

Evidence grades for product and supplier claims

These sit alongside the guide's own conventions (verified, interpretation, proposal, illustrative). They grade claims about products and suppliers, many of which come from community calls.

Verified

Confirmed against authoritative supplier documentation or a live configuration.

Corroborated

Reported by more than one independent source or session.

Reported

Raised in a community call or channel. Not yet confirmed.

Proposed

A design recommendation in this portal, not a finding.

Unresolved

An open question. Do not rely on it until answered.

Back to the contents · Suggest a correction (opens GitHub)

Part 4 · Assure and governChapter 4.2 · 21 of 76

Assure and govern · data and information governance

AI governance is data governance with actions attached

Whatever an AI tool can reach, it can surface, summarise and act on. Data and information governance sets what it should reach, and treats what it produces as records.

Ownership and accountabilityEvery source has an information asset owner; the senior information risk owner accepts residual risk.

Classification and handlingClassification travels with content into prompts, indexes and outputs.

Data protectionUK GDPR lawful basis, DPIA, transfers, rights and service-level privacy notices.

Quality and authorityGrounding sources are authoritative, current and fit for the decision they inform.

Metadata and discoverabilityTitles, filenames and metadata can surface through AI. Discovery is not access.

Records and retentionPrompts, outputs, transcripts and agent logs are records with retention rules.

Access and permissionsLeast privilege, oversharing clean-up, delegated and shared contexts tested.

Sharing and reuseData-sharing agreements, data boundaries and trusted reuse such as data matchmaking.

Data controls at each lifecycle stage

StageData and information controlWhere it is recorded
1 DiscoverIdentify the information assets involved and their ownersT1 Opportunity Brief
2 CheckReuse existing datasets, catalogues and indexes before creating new onesT2 Duplication Check
3 ClassifyClassification, personal data and special category dataT3 Classification
4 AssessDPIA, lawful basis, transfers and subprocessors, data quality and authorityT4 Data Readiness · T5 Supplier
5 DesignMetadata, index scope, permissions, connectors and retention rulesT7 Agent Design Record · data-source register
6 PilotMinimisation, test data, consent and privacy noticesT8 Pilot Plan
10 OperateAccess reviews, DLP, records of AI outputs and agent logsMonitoring and assurance record
11 ReassessNew sources, connectors or locations trigger reviewT11 Change-Impact Assessment
12 RetireDisposal, records transfer and revoked accessT12 Retirement Record

What AI creates is information too

Prompts and instructions

Retain where they shaped a decision or a published output; organisational prompts are versioned products.

AI outputs

Treat as drafts until a person verifies them; keep the verified version with an AI involvement statement.

Transcripts and recordings

Consent notice, retention period and deletion route set before recording starts.

Agent action logs

Keep for audit and incident review; owner can read them.

Indexes and embeddings

Derived copies of source data; inherit the source classification and retention.

Evaluation data

Test sets and results are evidence for the decision record.

Frameworks this aligns with

Data Ethics Framework ↗ (opens an external website)UK government framework for using data responsibly in policy and services.

Government Data Quality Framework ↗ (opens an external website)Principles and practices for assessing and improving data quality.

ICO guidance on AI and data protection ↗ (opens an external website)How UK GDPR applies to AI, including fairness, transparency and DPIAs.

Northern Ireland draft AI Strategy ↗ (opens an external website)Data governance is one of the eight proposed principles.

Scotland’s AI Strategy 2026–2031 ↗ (opens an external website)Data framed as a public asset, with a data matchmaking pilot.

Seen in practice

Kingston Council ran a DPIA on data flows, storage and access before going live with AI note-taking, and updated its privacy notice. Hartlepool publishes a service-level privacy notice for the same kind of tool. Belfast City Council put data and AI governance under one Data & Analytics Board.

Back to the contents · Suggest a correction (opens GitHub)

Part 4 · Assure and governChapter 4.3 · 22 of 76

Architecture of trust

Governing AI agents as digital workers

A continuous cycle that governs AI agents as digital workers to deliver trusted public services and outcomes. The agent at the centre is purpose driven, accountable, supervised, transparent, secure and ethical. Trust at every step: right agent, right authority, right outcome.

1

Identity

Know the agent
  • Define purpose and role
  • Register and record
  • Assign owners
2
Capability
What it can do
  • Define capabilities
  • Tools, data, knowledge
  • Evaluate and test
3
Authority
What it is allowed to do
  • Set boundaries
  • Permissions & limits
  • Approval thresholds
4
Policy & guardrails
How it must operate
  • Policies & rules
  • Ethics & responsible AI
  • Guardrails & constraints
5
Approval
Authorise to operate
  • Risk assessment
  • Assurance sign-off
  • Go-live approval
6
Execution
Act within context
  • Perform tasks
  • Interact with users
  • Create outputs
7
Evidence & traceability
Prove what happened
  • Logging and audit trails
  • Data provenance
  • Decisions explained
8
Audit & assurance
Check and challenge
  • Monitor & review
  • Independent assurance
  • Compliance checks
9
Learning & improvement
Get better over time
  • Feedback loops
  • Model & process updates
  • Continuous improvement
10
Exit & retirement
End safely and cleanly
  • Revoke access
  • Archive & retain records
  • Decommission & review

Guiding principles

Human accountability Safety, security and privacy Lawful, ethical and fair Transparency and explainability Inclusive and accessible Proportional and risk-based Evidence based and measurable Sustainable and value for money

Foundation standards

Service Standard Technology Code of Practice AI Playbook & Responsible AI Data & AI Ethics Framework Security & Privacy Standards ATRS & Transparency

Outcomes for public services

Better servicesSimpler, faster and more responsive

Trusted decisionsFair, consistent and well governed

Efficient operationsAutomate safely, reduce burden

Empowered workforceAugmented capability, not replaced

Improved outcomesBetter impact for citizens, business and communities

Stronger public trustAccountable, transparent and responsible AI use

Enabling foundations — built in, not bolted on

Security by design Protect systems, data and users

Privacy by design Respect and protect personal data

Human in control Meaningful oversight at all times

Impact aware Consider equality, ethics and public impact

Transparent Open about how and why agents are used

Measured & monitored Track performance, risk and value

Continuously improved Learn, adapt and evolve responsibly

Back to the contents · Suggest a correction (opens GitHub)

Part 4 · Assure and governChapter 4.4 · 23 of 76

Enterprise AI governance operating model

Nine layers, one accountable line of sight

A layered model for governing AI agents as digital workers to deliver trusted public services. Each layer answers one question — from why we do it, to the impact we create together.

Strategy & public value

Set direction, objectives and principles aligned to mission, public value and national priorities.

Why we do it
Policy & governance

Establish policies, roles, responsibilities, decision rights and governance structures.

Who is accountable and how
Standards & guidance

Apply mandatory standards and guidance (Service Standard, TCoP, AI Playbook, Responsible AI, data and security standards, ATRS).

What good looks like
Assurance & risk management

Identify, assess, mitigate and monitor risks across the AI lifecycle. Assure compliance, safety, ethics and performance.

How we manage risk and quality
Platforms & foundations

Provide secure, scalable, reliable platforms, data foundations, tools, shared services and reusable components.

What we build on
Services & solutions

Design and deliver AI-enabled services and solutions that solve real problems and meet user needs.

What we deliver
AI agents as digital workers

Managed AI agents with defined purpose, capabilities, authority, guardrails, tools, data access and human oversight.

Who or what performs the work
Users & workforce

Civil servants, partners and operational teams who work with AI agents, provide feedback and maintain oversight.

Who uses and oversees
Public outcomes & impact

Better services, fairer decisions, improved efficiency and outcomes for citizens, businesses and communities.

The impact we create together

Enablers across all layers

People & capabilityRoles, skills, training, culture and change enablement.

Security & privacySecure by design, privacy by design, data protection and resilience.

Ethics & inclusionFairness, equality, accessibility, human rights and inclusion.

Transparency & ATRSTransparency records, explainability, traceability and open publication where appropriate.

Data & knowledgeQuality data, governance, lineage, provenance and knowledge management.

Technology & architectureInteroperability, open standards, cloud first, modular and secure architectures.

Continuous improvementMeasure, learn, iterate and improve services, models and operations.

Outcomes for citizens, businesses and society

  • Accessible, inclusive and easy to use services
  • Safe, secure and trustworthy use of AI
  • Fairer decisions and improved experiences
  • Better outcomes, efficiency and productivity
  • Value for money and responsible use of resources
  • Stronger public trust, legitimacy and democratic accountability
Our commitment: governed AI agents that are safe, effective and accountable — empowering our workforce to deliver better public services.

Back to the contents · Suggest a correction (opens GitHub)

Part 4 · Assure and governChapter 4.5 · 24 of 76

Key risks & mitigations matrix

Identify early. Control appropriately. Monitor continuously. Improve relentlessly.

Proactive risk management so AI agents operate safely, lawfully and effectively as digital workers. Human accountability remains non-delegable. Residual ratings assume the controls listed are in place and effective.

1. Inaccurate outputs (hallucination)

Residual: Low

Description & impact
  • Agent generates incorrect, fabricated or misleading information.
  • Poor decisions, loss of trust, reputational and service impact.
Examples
  • Factual errors
  • Fabricated citations
  • Incorrect summaries
  • Outdated information
Key mitigations / controls
  • Grounding with trusted sources
  • Retrieval + citation requirements
  • Confidence thresholds
  • Human review for high-risk outputs
Assurance & monitoring
  • Output evaluation
  • Spot checks & sampling
  • User feedback loops
  • Red-teaming
Well controlled with evidence of accuracy and oversight.

2. Data privacy & leakage

Residual: Low

Description & impact

Sensitive or personal data is exposed, misused or shared inappropriately.

Breach of privacy law, fines, individual harm, loss of trust.

Examples
  • Prompt injection
  • Unintended data exposure
  • Insecure data handling
  • Over-broad access
Key mitigations / controls
  • Data classification & handling rules
  • Least privilege access
  • Data masking & filtering
  • Secure environments & logging
Assurance & monitoring
  • Access monitoring
  • Data loss prevention (DLP)
  • Privacy impact reviews
  • Audit trails
Strong controls and continuous monitoring in place.

3. Automation bias & over-reliance

Residual: Medium

Description & impact

Users over-trust agent outputs and fail to challenge or validate them.

Poor decisions, unintended harm, reduced critical thinking.

Examples
  • Acceptance without review
  • Over-trust in suggestions
  • Deskilling of staff
  • Suppressed escalation
Key mitigations / controls
  • Human-in-the-loop for key decisions
  • Clear UI prompts & explanations
  • Challenge culture & training
  • Escalation pathways
Assurance & monitoring
  • User behaviour analytics
  • Review rates
  • Escalation metrics
  • User training effectiveness
Managed through oversight and user behaviour.

4. Over-authorisation & scope creep

Residual: Medium

Description & impact
  • Agent given more authority or access than necessary or approved.
  • Unauthorised actions, data misuse, policy breach, financial loss.
Examples
  • Excessive permissions
  • Unapproved action types
  • Gradual expansion
  • Weak change control
Key mitigations / controls
  • Least privilege & role-based access
  • Capability versus authority separation
  • Change management process
  • Approval for new actions
Assurance & monitoring
  • Permission reviews
  • Access recertification
  • Change audit logs
  • Control effectiveness testing
Requires ongoing review and re-authorisation.

5. Bias & fairness issues

Residual: Medium

Description & impact

Agent outputs unfair, discriminatory or disproportionate results.

Inequitable outcomes, legal challenge, reputational harm, loss of legitimacy.

Examples
  • Biased training data
  • Proxy discrimination
  • Unequal performance
  • Lack of representation
Key mitigations / controls
  • Fairness assessments
  • Diverse data & testing
  • Bias detection tools
  • Human oversight & appeal
Assurance & monitoring
  • Fairness metrics
  • Outcome monitoring
  • Equality impact reviews
  • Independent assurance
Continuing evaluation reduces but does not eliminate risk.

6. Security threats & misuse

Residual: Medium

Description & impact

Agent exploited, manipulated or used maliciously by attackers.

System compromise, data theft, service disruption, fraudulent actions.

Examples
  • Prompt injection
  • Model exploitation
  • Adversarial attacks
  • Tool misuse
Key mitigations / controls
  • Secure by design
  • Input validation & guardrails
  • Threat modelling
  • Anomaly detection
Assurance & monitoring
  • Security monitoring
  • Pen testing & red teaming
  • Vulnerability scans
  • Incident response testing
Risk managed with robust security controls.

7. Model drift & performance degradation

Residual: Low

Description & impact

Agent performance declines over time as data, context or patterns change.

Reduced accuracy, poor outcomes, inefficiency, user dissatisfaction.

Examples
  • Changing data patterns
  • Outdated knowledge
  • Domain shifts
  • Feedback ignored
Key mitigations / controls
  • Continuous evaluation
  • Monitoring & alerting
  • Retraining & updates
  • Feedback integration
Assurance & monitoring
  • Performance dashboards
  • Drift detection
  • KPI tracking
  • Periodic revalidation
Active monitoring keeps risk low.

8. Accountability & governance gaps

Residual: Low

Description & impact

Unclear ownership, missing controls or weak governance arrangements.

Unclear responsibility, compliance breach, regulatory scrutiny, loss of trust.

Examples
  • No named owner
  • Weak oversight
  • Missing documentation
  • Poor auditability
Key mitigations / controls
  • Named owners & RACI
  • Governance framework
  • Mandatory documentation (ATRS)
  • Clear policies & escalation
Assurance & monitoring
  • Governance reviews
  • Audit & compliance checks
  • ATRS records reviewed
  • Board reporting
Strong governance minimises risk.

Foundational principles that reduce all risks

Human accountability People remain answerable.

Lawful & ethical use Compliant, fair and responsible.

Transparency & explainability Understandable actions and outputs.

Security & privacy by design Protect data and systems.

Inclusive & equitable Designed for all users and groups.

Measure & improve Evidence based continuous improvement.

Monitor & adapt Stay responsive to change.

Document & record Keep clear, complete and auditable records.

Back to the contents · Suggest a correction (opens GitHub)

Part 4 · Assure and governChapter 4.6 · 25 of 76

The full catalogue schema — eighteen fields

A single source of truth for AI agents: discoverable, governed and actionable. The four-group record above is the minimum; this is the full blueprint, with an illustrative value for each field.

WhyVisibility, control, safe adoption and reuse WhoOwners, users and support WhatPurpose, function and behaviour WhereData sources, systems and environment HowControls, assurance and monitoring Cost & valueUsage, cost and outcomes

1. Identity Agent ID (unique) · Name · Description · Version · Status e.g. AG-2026-015

2. Purpose & function Purpose · Primary functions · Typical tasks · Intended outcomes e.g. Summarise policy documents and answer internal questions

3. Ownership Business owner · Creator / builder · Technical owner · Information owner · Risk owner · Service owner e.g. Policy Team Lead

4. Users & audience Primary users · User groups · Audience size · Geographies / organisations e.g. 250 staff in 2 departments

5. Platform & model Platform (e.g. M365, Copilot Studio) · Model(s) used · Model version · Region / residency e.g. GPT-4o via Microsoft 365 Copilot

6. Data & information sources Source systems · Information assets · Grounding method (RAG, connectors) · Data sensitivity e.g. SharePoint libraries, Policy DB, HR system

7. Access & permissions Access model · Authentication · Permissions set · Least privilege confirmed e.g. Entra ID group based access

8. Sharing & distribution Sharing scope (personal / team / department / cross-gov) · Channels / locations · Public-facing? Y/N e.g. Teams tab, shared with 3 teams

9. Risk & criticality Risk category · Criticality (low/med/high) · Impact areas · Consequences e.g. Medium — operational impact

10. Controls & assurance Guardrails / controls · Human oversight points · Testing completed · Assurance status e.g. Human review before action

11. Usage & performance Usage level · Key metrics · Quality indicators · Last reviewed e.g. 1,250 queries / month, 92% helpful

12. Cost & consumption Licensing / pricing model · Tokens / credits used · Estimated monthly cost · Cost owner e.g. Pay-as-you-go, £320 / month, IT

13. Integrations & dependencies Connected systems · APIs / connectors · Dependencies · Third parties e.g. Graph API, Power Automate

14. Monitoring & observability Logging enabled · Monitoring tools · Alerts & thresholds · Incident contact e.g. Purview logs, alerts via Sentinel

15. Change & release Change process · Release history · Last updated · Next review date e.g. v1.2 — 12 Aug 2026, review 12 Nov 2026

16. Support & operations Support model · Support owner · SLA / hours · Escalation path e.g. IT Service Desk P1–P3, 24×7

17. Benefits & value Expected benefits · Measured outcomes · Baseline / target · Value owner e.g. Saves 10 hrs/week, faster responses

18. Retirement & archiving Retirement criteria · Data retention · Archive location · Decommission plan e.g. Retire if inactive 180 days

Status

Active In review Planned Retired Draft

Quick filters

By owner By department By status By risk By platform By data sensitivity

Governance at a glance

Policy & standardsAligned to organisational AI policy, standards and acceptable use.

Approval & publishingClear gates for publication and sharing at each tier of use.

Ongoing reviewRegular reviews of risk, performance, cost and relevance.

Discovery & reuseEasily discoverable to avoid duplication and share good practice.

Audit & accountabilityTraceable ownership, decisions, changes and evidence.

Agent lifecycle overview
  • 1. Idea — Identify need or opportunity
  • 2. Build — Create and test in a safe space
  • 3. Share — Share with team or small group
  • 4. Assess — Evaluate risk, quality and value
  • 5. Approve — Approve for wider use
  • 6. Operate — Monitor, support and optimise
  • 7. Review — Regular review of performance and risk

Three registers, working together

Integrated, not duplicated. Each serves a distinct purpose and links to the others for discovery, control and reuse — over one set of shared foundational services. Single source of truth · integrated, not duplicated · accessible and discoverable · governed and accountable.

Information asset register

Maintain authoritative inventory of information assets for governance, compliance and risk management.

Contains

Policies, standards & procedures Documents & records Data sets & databases Forms & templates Knowledge bases Reference data Expert directories

Key attributes

Owner & accountability · Classification & sensitivity · Retention & disposal · Legal & regulatory · Quality & governance

Data catalogue

Make data easy to find and understand to promote reuse, reduce duplication and improve decision-making.

Contains

Data sets & data products APIs & data services Metrics & KPIs Data dictionaries & models Knowledge bases Lineage & relationships Usage & popularity

Key attributes

Data definition · Owner & steward · Lineage & relationships · Quality & freshness · Access & usage

Agent register / catalogue

Register and govern AI agents and agentic capabilities to ensure safe, accountable and effective use.

Contains

Agents & agent teams Purpose & use cases Intended users & audience Data sources & tools Risk & criticality Controls & assurance Lifecycle & status

Key attributes

Agent owner & creator · Purpose & scope · Data sources & tools · Risk & criticality · Lifecycle & review

Shared foundational services

Identity & access management Security & privacy controls Metadata & standards (terms, definitions) Integration & APIs (data & systems) Audit & monitoring (usage, changes) Quality management (validation, rules) Search & discovery (federated)

How they work together

1. Create / ingest — New information, data or agents are identified and registered

2. Link & relate — Relationships are captured between assets, data and agents

3. Use & leverage — People and agents find and use trusted sources and capabilities

4. Monitor & assure — Activity, quality, risk and compliance are monitored across the landscape

5. Improve & evolve — Feedback drives better sources, smarter agents and stronger controls

The observability control tower

Real-time visibility of agents, usage, performance, risk and value — insight into every agent, control where it matters, confidence in outcomes. These are the panels the register needs to earn its keep.

Agents by status Active · In review · Retired

Usage over time Total agents against active agents, last 30 days

Top agents by interactions Ranked list with trend sparklines

Risk distribution Very high · High · Medium · Low

Agents by owner type Individual · Team · Department · Enterprise

Spend by agent Cost per agent, highest first

Quality & reliability Success rate · Response time · Feedback score · Error rate

Sharing & discovery Shared · Public to organisation · Cross-organisation

Alerts & needs attention Agents with no active owner · Breached sharing thresholds · Not reviewed in 60+ days · Rising error rate

Recommended actions Assign ownership and confirm purpose · Review sharing scope · Confirm continued value and risks · Validate cost and usage patterns

Filter by dimension

Owner Department Agent type Platform / tool Data sensitivity Risk level Lifecycle status Tags / capability Geography

Back to the contents · Suggest a correction (opens GitHub)

Part 4 · Assure and governChapter 4.7 · 26 of 76

Internal ATRS registry · AI agent transparency record

A transparency record for every agent used inside the organisation

Based on the principles of the Algorithmic Transparency Recording Standard. Purpose: to ensure transparency, accountability and responsible use of AI agents by recording key information about their purpose, authority, data, risks, controls and lifecycle. The registry integrates with enterprise systems for identity, security, monitoring, risk, audit and knowledge management.

1

Agent identity

  • Agent name
  • Unique identifier
  • Version
  • Owners (SRO, service, technical)
  • Sponsor / business unit
  • Service supported
2
Business context & purpose
  • User needs addressed
  • Problem statement
  • Intended outcomes
  • Users and stakeholders
  • Public impact assessment
  • Success measures
3
Authority & permissions
  • Actions permitted
  • Actions prohibited
  • Decisions supported
  • Escalation triggers
  • Human approval requirements
  • Delegated authority level
4
Data & knowledge
  • Data sources & types
  • Data classification
  • Knowledge sources
  • Retention & disposal
  • Provenance & lineage
  • Data sharing & flows
5
Responsible AI considerations
  • Fairness & bias assessment
  • Equality & inclusion impact
  • Accessibility considerations
  • Ethical considerations
  • Human oversight approach
  • Potential harms & mitigations
6
Assurance & controls
  • Security review
  • DPIA completed
  • Risk assessment
  • Evaluation results
  • Red-teaming
  • Compliance checks
  • Approval & sign-off
7
Operational management
  • Monitoring & logging
  • Performance metrics
  • Incident management
  • Alerting & thresholds
  • Kill switch / disable
  • Rollback process
  • Support & escalation
8
Lifecycle & status

Lifecycle stage (proposed, pilot, live, restricted, suspended, retired)

  • Start date
  • Review frequency
  • End of life plan
  • Decommission date

Benefits

  • Builds trust and accountability
  • Supports assurance and audit
  • Enables oversight and control
  • Promotes continuous improvement
  • Aligns with public sector standards

Outputs & uses

Central inventory of AI agents Transparency for oversight Evidence for audits & assurance Risk management & reporting Informed decision making Public sector standards compliance Learning & improvement

Record quality: keep the record accurate, up to date and complete. Review at defined intervals or when significant changes occur.

Back to the contents · Suggest a correction (opens GitHub)

Part 4 · Assure and governChapter 4.8 · 27 of 76

Failure & recovery loop

Learn fast. Fix the issue. Prevent recurrence. Share the lesson.

Turning errors into stronger systems and better outcomes. Nine stages, and the ninth feeds the first: every failure is a source of learning. Work a live issue through the loop and record where you are.

TransparencySurface issues quickly AccountabilityOwn and act on problems LearningCapture and share lessons Continuous improvementStrengthen continuously ProportionateRespond to risk and impact

0 of 9 stages recorded

Stage 1 of 9

Detect & identify

Identify that something has gone wrong.

Errors, unexpected outputs User feedback Monitoring & alerts Quality checks

Typical triggers

Inaccurate or incorrect outputs

Outdated or incorrect data sources

Prompt or model misbehaviour

User confusion or misuse

System or integration failure

Policy or compliance breach

Success indicators
  • Issues detected early
  • Risks contained
  • Root causes addressed
  • Controls strengthened
  • Lessons shared and applied
  • Fewer repeat incidents
  • Improved outcomes and trust
Enablers
Leadership sponsorship
Clear roles & responsibilities
Policies & standards
Quality data & governance
Security & privacy by design
Observability & monitoring
Skilled teams & training
Open feedback culture
Tools & automation support

Back to the contents · Suggest a correction (opens GitHub)

Part 5 of 12 · 6 chapters

Operate and oversee

Part 5 · Operate and overseeChapter 5.1 · 28 of 76

AI governance control tower · illustrative dashboard

AI agent portfolio control tower

Govern. Assure. Monitor. Improve. Every figure below is illustrative, showing the dashboard a governed portfolio produces — last updated 20 May 2025 09:30.

128

AI agents registered

↑ 12 this month

87

Live in production

↑ 7 this month

45

In testing / pilot

↓ 3 this month

5

Suspended

— No change

12

Retired

↑ 2 this month

116

ATRS records published

↑ 10 this month

Agents by lifecycle stage

128 total

Proposed 8 6%

Live 87 68%

Pilot / testing 45 35%

Suspended 5 4%

Retired 12 9%

Assurance status

128 total

Assured 72 56%

Partially assured 28 22%

Not assured 18 14%

Not required 10 8%

Risk overview

High risk 12 9%

Medium risk 34 27%

Low risk 62 48%

Very low risk 20 16%

12 agents are high risk and require action.View risk register →

Agents by domain

Citizen services 32 25%

Corporate services 24 19%

Policy & strategy 18 14%

Operations 16 13%

Finance 14 11%

HR & people 12 9%

Other 12 9%

AI agent register

View all agents →

AGENT NAME SERVICE / DOMAIN PURPOSE OWNER LIFECYCLE RISK ASSURANCE ATRS RECORD LAST REVIEWED NEXT REVIEW

Policy Briefing Assistant

Policy & Strategy

Supports policy analysts by summarising key documents and drafting briefing notes.

Sarah Johnson (Director)

Live
High
Assured
ATRS-00045

15 May 2025

15 Aug 2025

Casework Triage Agent

Citizen Services

Prioritises and routes incoming casework to the right teams.

Mark Patel (Head of Ops)

Live
Medium
Partially assured
ATRS-00032

10 May 2025

10 Aug 2025

Finance Anomaly Detector Finance Detects unusual transactions and supports investigation. Alex Kim (Finance Lead) Live Medium Assured ATRS-00028 08 May 2025 08 Aug 2025

HR Policy Advisor HR & People Provides guidance on HR policies and employee queries. Lucy Bennett (HR Manager) Pilot Low Assured ATRS-00061 12 May 2025 12 Aug 2025

IT Support Copilot

Corporate Services

Assists IT staff with troubleshooting and knowledge retrieval.

Tom Edwards (IT Manager)

Pilot
Low
Not required
—

18 May 2025

18 Aug 2025

Comms Drafting Agent Corporate Services Drafts communications and public messages. Nina Shah (Comms Lead) Suspended High Partially assured ATRS-00017 02 May 2025 Review pending

46Data sources connected

27High sensitivity data sources

96%Agents have named owners

100%DPIA completed (in scope)

94%Security reviews up to date

85%Agents have defined success metrics

65%Agents have sustainability assessment

Control tower navigation

Overview
AI agents
ATRS registry
Assurance
Risks
Incidents
Reports
Analytics
Settings
Help

Quick actions

Register new agent
Create ATRS record
Request assurance
Report an issue

Back to the contents · Suggest a correction (opens GitHub)

Part 5 · Operate and overseeChapter 5.2 · 29 of 76

Responsible AI · operate and improve

Approval stays current only while the evidence does

Monitoring, incidents, the change watchlist, reassessment and retirement, run as one learning loop. The Portfolio control tower shows the estate view; this chapter defines what feeds it.

Dashboard domains

DomainWhat it showsFed byAct when
Adoption and usageWho uses which capability, and how often, by group.Tool, use-case registersUptake is uneven across groups or roles.
Cost and consumptionSpend by tool, team and use case against caps.Tool, benefits registersA cap or alert is hit, or cost grows faster than use.
Outcome quality and valueVerified outcomes, not prompt counts or time saved alone.Benefits registerValue evidence is missing at review.
Correction and reworkReview effort and how much output is edited or discarded.Benefits registerRework cancels the time saved.
Accessibility and inclusionBarriers reported, adjustments supported, alternatives used.Incident, use-case registersA barrier is reported without an alternative.
Risk, incidents and near missesIncidents by tool, model, use case and control.Incident registerRepeat incidents on one control.
Sustainability indicatorsSupplier environmental evidence and usage indicators.Tool, supplier registersEvidence is missing or usage is disproportionate.
Control effectivenessWhether controls worked when tested or triggered.Incident, review registersA control failed or was bypassed.
Change and deprecation exposureLive use affected by upcoming supplier changes.Change, model registersA deprecation lands inside the review window.
Support demand and recurring issuesTickets and repeat questions by tool.Support and service recordsThe same issue recurs across teams.

No figures are shown. Each domain is defined so live data can be connected without redesign.

Changes that trigger reassessment

Model

A new, changed or automatically selected model.

Partial or full

Supplier

New subprocessor, location, terms or ownership.

Full

Data

New sources, connectors or classifications.

Partial or full

Permission

Wider access, new identities or delegation.

Full

Action

New tools, actions, schedules or agent links.

Full

Interface

Changed terminology, layout or defaults.

Guidance update

Cost

New pricing, metering or consumption pattern.

Partial

Policy

New law, policy or guidance.

Partial or full

Change and deprecation watchlist

17–20 Nov 2026

Microsoft IgniteSeveral preview items in the service atlas are likely to change. Refresh entries after the event.

Proposed

Ongoing

Staggered Copilot interface rolloutUsers see different interfaces at once. Keep guidance by release channel and show preview versus general availability.

Reported

Each release

Default-enabled featuresNew features can arrive switched on. Check admin defaults after each release.

Reported

Open

Claude models reached through MicrosoftConfirm subprocessor terms and whether processing sits inside the data boundary.

Unresolved

Open

Microsoft UK data processing commitmentConfirm in your admin centre before relying on it.

Unresolved

Ongoing

Indexed metadata exposureTitles, filenames and metadata can surface despite permissions. Re-test indexing on change.

Reported

Sep 2026

Cowork cost and securityConcerns raised about credit consumption and security. Set caps before any pilot.

Reported

Governance as a continuous learning system

  1. 1. Pilot

    • Define scope and success measures
    • Test with real users
    • Put controls in place
    • Set stop criteria
  2. 2. Evidence

    • Gather quantitative and qualitative data
    • Assess outcomes, quality and value
    • Capture accessibility, cost and risk
    • Record unintended effects
  3. 3. Decision

    • Approve, approve with conditions, extend, restrict or stop
    • Record the rationale
    • Update registers
    • Tell users
  4. 4. Operation

    • Deploy and support users
    • Monitor controls and compliance
    • Provide guidance and training
    • Maintain human oversight
  5. 5. Monitoring

    • Track usage and adoption
    • Monitor cost and consumption
    • Identify risks and anomalies
    • Measure outcomes and value
  6. 6. Incident, change or learning

    • Respond to incidents and near misses
    • Investigate and resolve
    • Capture lessons
    • Watch product, model and policy changes
  7. 7. Reassessment

    • Assess the impact of change
    • Update the risk and control position
    • Refresh approvals if needed
    • Decide: continue, restrict or suspend
  8. 8. Retire

    • Disable access and actions
    • Remove data and integrations
    • Inform users
    • Close records and capture lessons

Turning experience into impact

Better decisionsBased on real outcomes Stronger controlsInformed by actual risks Greater valueFor users and public services More inclusiveIdentifying and removing barriers Smarter spendingUnderstanding cost and benefit Shared learningReusable patterns and guidance

Report an AI concern, incident or near miss

Report anything that went wrong or nearly did: a wrong output that was acted on, an exposed file, an unexpected action, a cost spike, an accessibility barrier. Record:

  • Tool, model and deployment route
  • Use case and what happened
  • Who was affected, and whether it was acted on
  • The control that should have prevented it
  • Corrective action and owner

Suspension and retirement

  1. Record the reason and the replacement.
  2. Tell users, with an accessible alternative.
  3. Disable agents and actions, and revoke access.
  4. Disconnect integrations and dispose of data.
  5. Update registers and remove supplier access.
  6. Record residual risks and lessons; approve closure.

Back to the contents · Suggest a correction (opens GitHub)

Part 5 · Operate and overseeChapter 5.3 · 30 of 76

Agent journey

From idea to retirement, with a gate at every stage

An agent that helps one person and an agent that decides a citizen's case are not the same thing, and should not carry the same controls. Eight stages, each with a gate question, the control that applies, the role accountable and the outcome expected.

Stage 1 of 8 Personal / low risk

Idea & discovery

What happens
  • Identify a need or opportunity
  • Explore feasibility
  • Check existing solutions and policies
Gate 1 — decision point
Is there a clear need and no existing solution?
Control applied

Security & identity — least privilege, access control and monitoring

Key accountability

Creator — builds and tests the agent; owns initial purpose

Outcome at this stage
Better understanding of needs and opportunities
Success factors across the journey

Clear purpose and value Right controls at the right time User engagement and feedback Sustainable cost model Continuous learning and improvement Strong community and reuse Better public value outcomes

Personal experiment to enterprise service

A managed progression that balances innovation, control, risk and value. Agents evolve from personal productivity tools to trusted, supported services — and ownership moves with them.

1 Idea & personal experiment Individual creator Low risk · personal

2 Share & team pilot Team data owners Low to moderate risk

3 Managed team agent Named team owner Moderate risk

4 Departmental capability Departmental owner Moderate to high risk

5 Operational service Service owner High risk

6 Enterprise service Business / enterprise owner High risk · high impact

7 Cross-government capability Cross-government sponsor & owners High impact · strategic

The Copilot-to-agent capability spectrum

A continuum of assistance and autonomy — from individual productivity to enterprise agentic services. Purpose: enable people to work better today, orchestrate intelligent workflows, and deliver trusted, efficient, inclusive public services.

Tier 2 of 7 · Microsoft 365 Copilot

M365 Copilot

AI in your apps to summarise, draft, analyse and explore

TYPICAL USERSIndividual

AUTONOMY LEVELHuman-guided

SCOPEPersonal

EXAMPLESWord drafting · Excel analysis · Meeting summary

GOVERNANCEOrg policies

Avoiding the “access database moment”

How a small, helpful agent becomes business critical without visibility or control. Small agents can create big value — and big risk. Visibility, ownership and governance early in the journey prevent unmanaged tools from becoming unmanageable dependencies.

1 Personal experiment Low visibility An individual creates an agent to solve a personal problem or save time.

2 Shared with team Low to moderate visibility Shared with colleagues or a small team to help with common tasks.

3 Wider adoption Rising visibility More teams discover it and start using it organically.

4 Dependence builds Rising to high visibility Critical work starts to rely on the agent and its outputs.

5 Business critical High visibility Used across the department or service. Business processes become dependent on it.

6 Invisible criticality Gap in visibility No formal ownership, limited documentation, unknown controls, unknown costs.

7 High risk exposure Critical · unmanaged Risk of failure, data breach, inaccurate decisions or service disruption.

8 Event of failure Loss of trust and value Agent breaks, data changes or owner leaves — causing a real operational impact.

How to prevent it — early detection, ownership and routes to enterprise management

1. Discover early Agent inventory · Usage monitoring · Sharing visibility · Community reports

2. Require description Purpose & use case · Data sources · Support owner · Review dates

3. Assign ownership Business owner · Technical owner · Information owner · Risk owner

4. Assess & classify Risk & impact · Data sensitivity · Criticality score · Assurance needs

5. Control & assure Approved sources · Guardrails · Human oversight · Testing & validation

6. Monitor & review Usage & performance · Cost & consumption · Accuracy & quality · Incident tracking

7. Promote or remediate Move to managed platform (e.g. Copilot Studio) · Or retire / remediate

What goes wrong when it is left unmanaged

Shadow IT Inconsistent quality Lack of documentation Knowledge silos Unknown cost Compliance breach Loss of public trust

Back to the contents · Suggest a correction (opens GitHub)

Part 5 · Operate and overseeChapter 5.4 · 31 of 76

Agent employment lifecycle

Managing AI agents as governed digital workers

Agents go through a managed lifecycle — from intent to retirement — with controls, oversight and accountability at every stage. People remain accountable: agents augment human capability, they do not replace judgement, responsibility or accountability.

1

Recruitment (define need)

Human analogy: Job requisition & role definition

Identify the need, purpose and expected outcomes.

Key controls

  • Business case
  • User need
  • Impact assessment
  • Initial risk screen
Clear mandate and value case
2
Vetting & evaluation
Human analogy: Recruitment, screening & interviews

Assess capability, risks, data needs and alignment with policy and standards.

Key controls

  • Capability evaluation
  • Risk assessment
  • Data assessment
  • Policy alignment
  • Assurance readiness
Fit for purpose and safe to proceed
3
Onboarding & configuration
Human analogy: Induction & onboarding

Configure the agent, set permissions, guardrails and connect data and tools.

Key controls

  • Guardrails
  • Permissions
  • Data access
  • Tool access
  • Prompt / policy setup
Ready, secure and configured correctly
4
Probation (pilot)
Human analogy: Probation period

Operate in a controlled pilot with close human oversight and limited scope.

Key controls

  • Limited scope
  • Human-in-the-loop
  • Enhanced monitoring
  • Evaluation criteria
Validated in real-world conditions
5
Live operation (employment)
Human analogy: Active employment

Agent operates within its delegated authority to support people and processes.

Key controls

  • Performance KPIs
  • Quality checks
  • Bias & fairness review
  • Value realisation review
Measurable performance and value
6
Performance review
Human analogy: Performance appraisal

Regularly review performance, accuracy, impact, risks and value delivery.

Key controls

  • Model updates
  • Knowledge refresh
  • Prompt tuning
  • Control improvements
Improved capability and reduced risk
7
Retrain & improve
Human analogy: Training & professional development

Update knowledge, models, prompts and controls based on learning and feedback.

Key controls

  • Restrict authority
  • Increase oversight
  • Root cause analysis
  • Remediation plan
Risk managed and issues contained
8
Suspension / restriction
Human analogy: Suspension

Restrict or pause the agent if risks increase, rules are breached or performance drops.

Key controls

  • Restrict authority
  • Increase oversight
  • Root cause analysis
  • Remediation plan
Risk managed and issues contained
9
Retirement & offboarding
Human analogy: Resignation or termination

Retire the agent when no longer needed or superseded. Revoke access and archive records.

Key controls

  • Revoke access
  • Disconnect tools
  • Data cleanup
  • Archive agent artefacts
  • Stakeholder notification
Safe retirement and clean closure
10
Archive & audit
Human analogy: HR records retention & reference

Retain records for audit, transparency and learning in accordance with retention rules.

Key controls

  • Records archive
  • Audit trail retention
  • Lessons learned
  • Knowledge capture
Accountable, transparent and auditable

Applied throughout every stage

Continuous monitoring Human oversight Audit logging Policy compliance Security Privacy Safety

Capability is not authority

An agent may be able to do many things, but can only do what it is authorised to do. Capability is about potential; authority is about permission. Capability + authority + context = action — only when all controls are satisfied can the agent act.

What the agent can do — capability

Technical potential and skills: the full range of capabilities, knowledge, tools and actions the agent is built or enabled to perform.

Natural language understanding Tool use & API integrations Reasoning & problem solving Workflow orchestration & scheduling Data retrieval (RAG) Analysis & visualisation Code generation & automation Communication (drafting, email, chat) Content creation & summarisation Learning from feedback

Governance controls

Bridging capability to appropriate action.

Policies & rulesOrganisational policies, legal requirements and ethical guardrails define boundaries.

Delegated authorityExplicit permissions granted for specific tasks, data and systems.

Conditions & controlsContext, thresholds, approvals and risk checks must be satisfied.

Human oversightPeople remain in control of decisions and can intervene or override.

Monitoring & auditActions are logged, monitored and reviewed for assurance and compliance.

What the agent is allowed to do — authority

Permitted actions in the right context, within defined boundaries, for specific purposes and users.

Draft policy briefingApproved for Policy Team (sensitivity: OFFICIAL)

Send standard email responseApproved with template and review

Summarise documents up to OFFICIALApproved with data classification filter

Update internal tracker or systemApproved with role-based permissions

Escalate issue to human reviewerRequired when confidence below threshold

Why separate capability from authority

Reduces risk Limits exposure by ensuring agents cannot act beyond intent or policy.

Improves accountability Clear ownership and delegated authority for all actions.

Supports compliance Meets legal, regulatory and ethical obligations.

Enables agility Adjust permissions without rebuilding capabilities.

Builds trust Transparent controls and oversight build confidence.

Adapts to context Decisions and permissions can vary by role, risk and situation.

Back to the contents · Suggest a correction (opens GitHub)

Part 5 · Operate and overseeChapter 5.5 · 32 of 76

Public service impact cascade

How agents influence decisions and deliver public value

Seven links from capability to legitimacy — each with the elements that must hold and the risks that arise if they do not. Better governed AI agents lead to better decisions, better services, and better outcomes for everyone.

1

AI capability (agent)

The agent has defined capabilities, knowledge, tools and guardrails within its delegated authority.

Key elements

  • Purpose & scope
  • Data & knowledge
  • Tools & actions
  • Guardrails & policies
  • Monitoring in place

Risks at this stage

  • Inaccurate data
  • Bias in training
  • Over-broad scope
  • Weak guardrails
2
Civil servant (user)

Civil servants interact with the agent, using its insights, suggestions or automation to support their work.

Key elements

  • Trained & supported
  • Understood context
  • Human judgement
  • Challenge & review
  • Accountability clear

Risks at this stage

  • Over-reliance
  • Skill gaps
  • Poor prompting
  • Lack of challenge
3
Operational decision

The agent's output informs or automates parts of operational decisions and actions within the service process.

Key elements

  • Decision boundaries
  • Human oversight
  • Documentation
  • Compliance checks
  • Audit trail captured

Risks at this stage

  • Automation bias
  • Unclear accountability
  • Policy non-compliance
  • Incomplete records
4
Service outcome (internal)

Operational actions produce internal service results such as case resolution, guidance issued, or processing completed.

Key elements

  • Process efficiency
  • Accuracy & quality
  • Timeliness
  • Consistency
  • Cost effectiveness

Risks at this stage

  • Process error
  • Delay or backlog
  • Poor integration
  • Resource waste
5
Public experience (end users)

Citizens, businesses and communities experience the service through interactions, outcomes and communications.

Key elements

  • Accessible & inclusive
  • Fair & equitable
  • Clear & understandable
  • Safe & secure
  • Easy to interact with

Risks at this stage

  • Exclusion
  • Confusing comms
  • Privacy concerns
  • Unequal treatment
6
Public value & outcomes

Better outcomes are achieved for people and society, improving trust in services and contributing to public value.

Key elements

  • Better outcomes
  • Reduced burden
  • Improved equity
  • Economic value
  • Confidence & trust

Risks at this stage

  • Unintended harm
  • Widening inequity
  • Benefit not realised
  • Poor measurement
7
Trust & legitimacy

Transparent, accountable and responsible use of AI strengthens public trust and democratic legitimacy.

Key elements

  • Transparency
  • Accountability
  • Safety & ethics
  • Oversight & redress
  • Continuous improvement

Risks at this stage

  • Lack of transparency
  • Accountability gaps
  • Safety incidents
  • Erosion of trust

Enablers & controls across the cascade

Strong governance & ownership Policies & ethical use Security & privacy by design Quality data & provenance Monitoring & auditability Human oversight at every step Transparency (ATRS) Performance measurement Continuous improvement

Back to the contents · Suggest a correction (opens GitHub)

Part 5 · Operate and overseeChapter 5.6 · 33 of 76

Observatory & barometer

Published signals and a readiness assessment

The observatory tracks published signals of UK public-sector AI adoption; the model is the OECD.AI Policy Observatory, applied at home. Each figure carries its publisher and date. The barometer is your own assessment, stored on this device only.

Observatory — published signals

OECD.AI · 2026

125+

ATRS transparency records published. Mandatory reporting for central departments and ALBs since 2025.

Civil Service · 2026

252

ideas entered in the Civil Service AI & Data Challenge 2026 — a 160% increase, in its fourth year.

DSIT analysis · 2026

£68bn

AI investment pledged into the UK since January 2025.

delivery.ai.gov.uk · 2026

21 → 420

UK AI compute in ExaFLOPs: 2025 actual to 2030 commitment (from 2 in 2024).

DSIT analysis · 2026

200k

People studying AI-related higher-education programmes in the UK.

Heriot-Watt & GoLLM · 2025

1 in 3

Scottish councils piloting or using AI. The leading UK nation for local-government adoption.

Barometer — your readiness

Rate each dimension 0–4. Saved locally on this device; nothing leaves the page.

—
Rate the dimensions for a reading

0 of 8 dimensions rated

Strategy & leadership
Governance & assurance
Data readiness
Skills & multidisciplinary team
Security posture
Ethics & transparency
Delivery capability
Monitoring & evaluation

Back to the contents · Suggest a correction (opens GitHub)

Part 6 of 12 · 4 chapters

Fund and buy

Part 6 · Fund and buyChapter 6.1 · 34 of 76

Cost, value & investment · illustrative figures

Make every AI pound count

A continuous cycle that connects business value, cost efficiency and responsible use. Align, optimise, govern — then deliver and reinvest. Figures on this page are illustrative: replace them with your own contract and telemetry data.

1 Forecast & plan Understand demand, use cases, models and unit economics.2 Allocate & budget Allocate costs, set budgets and guardrails.3 Observe & measure Monitor usage, costs, quality and performance in real time.4 Analyse & evaluate Analyse spend, quality, outcomes and business value by use case.
5 Optimise & improve Optimise models, workloads, prompts and architecture.6 Report & transparency Report on spend, value, carbon and compliance.7 Reinvest or rebalance Reinvest in high-value initiatives or rebalance priorities.8 Govern & control Review policies, controls and risks. Adapt and evolve.

What we manage

Consumption costsTokens, credits, API calls, capacity, compute, storage, network

Technology costsModels, platforms, connectors, licences, infrastructure

Adoption costsTraining, change, champions, workflow redesign, support

Data & integration costsData preparation, quality, integration, access and governance

Assurance & risk costsSecurity, privacy, testing, human review, audits, incident response

Operational costsMonitoring, tooling, run support, maintenance, model updates

Transition & exit costsMigration, parallel run, data portability, vendor switching

Value we deliver

Better public servicesImproved outcomes for citizens

Productivity gainsTime saved, efficiency improved

Quality & consistencyFewer errors, higher quality outputs

Risk reductionStronger controls, fewer incidents

Transparency & trustClear oversight, explainable decisions

Sustainable impactLower waste, optimised resources, lower carbon

Key practices

  • Tag and track by use case
  • Set budgets and alerts
  • Right-size models and workloads
  • Reduce token waste
  • Automate policies and guardrails
  • Review value regularly
  • Collaborate across business, technology, finance and security
Who works together

Business owners Product & delivery Finance / FinOps Engineering & data Security & privacy Procurement & commercial Operations & service management

Unit economics: what one interaction costs, and what it returns

Total cost per interaction = token cost + infrastructure + tool and API cost + human effort + overhead. Multiply by volume for the period, and set it against the value delivered. The starting values are the source model's worked example.

Cost inputs
Cost per interaction

Input tokens 2,000 tokens @ £0.002 / 1K £0.0040

Output tokens 1,000 tokens @ £0.006 / 1K £0.0060

Tool / API calls External calls and connectors £0.010

Infrastructure Compute, storage and network allocation £0.0050

Human effort 5 minutes @ £1 / min £5.00

Overheads Operations, support and monitoring £0.500

Total cost per interaction £5.53

× 1,000 interactions in the period £5,525.00

Value per interaction

Total value £20.00

NET VALUE £14.48 RETURN 262%

Value to cost ratio 3.6× · value across the period £20,000.00. Human review time is usually the largest single line — design it down before changing model.

1
Workload / request

User or system initiates a task

Examples
  • Summarise document
  • Draft policy response
  • Code generation
  • Case classification
  • Data analysis
2
Tokens & model consumption

Tokens processed by the model

Components
  • Input tokens (prompt)
  • Output tokens (response)
  • Reasoning / tool use tokens
  • Cached tokens
  • Embedding tokens
3
Cost drivers

What drives the cost of the interaction

Cost elements
  • Model pricing per 1K tokens
  • Context length
  • Frequency & volume
  • Tool / API calls
  • Storage & compute
  • Network / egress
4
Quality & outcomes

Quality of output and outcome achieved

Quality factors
  • Accuracy & relevance
  • Hallucination rate
  • Completeness
  • Safety & compliance
  • User satisfaction
  • Rework rate
5
Human effort & rework

Human time to review, edit or redo

Effort factors
  • Review time
  • Editing / refinement
  • Additional research
  • Escalations
  • Rework / repeat runs
  • Approval steps
6
Business value

Value delivered to the organisation

Value types
  • Time saved
  • Productivity gain
  • Better decisions
  • Risk reduction
  • Improved quality
  • Citizen / service impact
Cost and value optimisation levers

Right modelMatch quality needs to the most cost-effective model

Prompt engineeringOptimise prompts to reduce tokens and improve outcomes

Context managementRight-size context, use retrieval and caching

Routing & policiesRoute to the best model or tier based on cost and risk

Caching & reuseReuse responses, embeddings and results

Evaluation & guardrailsMeasure quality and set thresholds to avoid waste

FinOps controlsBudgets, alerts, showback / chargeback and governance

KPIs to track

Cost per 1,000 tokens Cost per interaction Cost per outcome Quality score Rework rate Time saved ROI / value per £ Budget adherence Token utilisation efficiency

Six cost layers, each with an owner

Fair, transparent, actionable. Allocate costs across the organisation to drive accountability, optimise spend and demonstrate value. Without a named owner, a layer is not allocated — it is absorbed.

LAYER 1
Enterprise foundation
  • Core platform and shared capabilities enabling all AI use.
  • Azure / cloud baseline
  • Identity & security
  • Networking
  • Monitoring platform
  • Central services

OWNER CTO / enterprise Direct allocation by actual cost

LAYER 2
Shared platform services
  • Reusable services consumed by multiple products and teams.
  • AI platforms / services
  • Data platforms
  • Model hosting
  • Logging & observability
  • Shared tools & libraries

OWNER Platform / centre of excellence Cost pools and apportionment drivers

LAYER 3
Product / service
  • AI capabilities built into a product or service.
  • Feature development
  • Data preparation
  • Models & fine-tuning
  • QA & evaluation
  • Product support

OWNER Product / service owner Showback or chargeback to products

LAYER 4
Team / workload
  • Teams running workloads to deliver features or operations.
  • Training / inference
  • API calls / tokens
  • Experimentation
  • Environments
  • Team tooling

OWNER Team lead Team budgets and workload tagging

LAYER 5
Use case / consumer
  • A specific AI use case delivering value to a defined user group.
  • End-user interactions
  • Business process cost
  • Human-in-the-loop
  • Quality assurance
  • Change & adoption

OWNER Use case owner Usage-based allocation (tokens, calls, transactions)

LAYER 6
End user / beneficiary
  • Cost associated with delivery to end users or citizens.
  • Per user / transaction
  • Support & helpdesk
  • Feedback & outcomes
  • Accessibility support
  • Service impact

OWNER Business / service Unit cost per user or beneficiary

Allocation flow
1. Capture actual costs
2. Tag & classify costs
3. Allocate using defined rules
4. Report & analyse
5. Optimise & act
6. Value realised
What we gain
  • Clear visibility of total cost
  • Fair and consistent allocation
  • Informed decision making
  • Cost optimisation
  • Accountability and ownership
  • Evidence of value and return
Data & enablers
  • Consistent tagging & metadata
  • Usage & cost data feeds
  • Access & permissions
  • Cost ownership model
  • Policies & standards
  • Automation & integration
  • Data quality & validation
Governance & controls
  • Cost policies & guardrails
  • Budgeting & approvals
  • Monitoring & alerts
  • Audit & reconciliation
  • Roles & responsibilities

Visible costs are the tip of the iceberg

Licences, tokens and compute are usually budgeted and tracked, and are only a fraction of the total. The hidden lines below the waterline determine real value and long-term success.

Licences & subscriptions
Token / usage charges
Cloud compute & storage
Hidden costs — what is often overlooked

Data & information

  • Data quality & remediation
  • Information architecture
  • Permissions & access
  • Metadata & cataloguing

Integration & engineering

  • Connectors & APIs
  • Workflow redesign
  • Systems integration
  • Environment set-up

Security & compliance

  • Security controls
  • Privacy & DPIAs
  • Compliance & legal
  • Pen testing & assurance

Adoption & change

  • Training & enablement
  • Champions & local support
  • Communications
  • Workflow & culture change

Assurance & evaluation

  • Model evaluation
  • Bias, fairness & robustness
  • Human review effort
  • Audit & red teaming

Operations & support

  • Monitoring & observability
  • Incident response
  • Helpdesk & support
  • Continual improvement

FinOps & management

  • Cost monitoring & allocation
  • Forecasting & optimisation
  • Showback / chargeback
  • Demand management

Governance & risk

  • Policies & standards
  • Risk management
  • Third-party risk
  • Records & retention

Workforce & roles

  • Role evolution
  • Skills development
  • Capacity & backfill
  • HR & employee relations

Sustainability

  • Energy & carbon impact
  • Efficient model usage
  • Sustainable operations
  • Reporting & metrics

Transition & exit

  • Parallel running
  • Data portability
  • Supplier exit costs
  • Migration & decommission

Value realisation

  • Benefit tracking
  • Outcome measurement
  • Realisation management
  • Continuous feedback

Total cost of AI = technology + data + people + process + assurance + operations + commercial. Budget for the full cost. Manage it actively. Realise the value.

What drives hidden costs

ComplexityMany moving parts across technology, data and people

ChangeWorkflows, roles and ways of working must adapt

Risk & assuranceHigher risk requires stronger controls and evidence

Scale & usageMore users, data and interactions increase ongoing costs

GovernancePolicies, oversight and reporting add essential investment

Make the hidden visible
Map all cost categories
Track usage & value
Allocate transparently
Optimise continuously
Review & adapt

Beyond licences: adoption as an investment line

AI adoption is a long-term investment across technology, people, process, data and assurance. Sustainable value comes from funding the whole system, not the licence.

1. Technology
  • Licences & subscriptions
  • Tokens / consumption
  • Models & platforms
  • Infrastructure & cloud
  • Tools & integrations
2. People & adoption
  • Training & skills
  • Champions & support
  • Communications
  • User research
  • Protected time
3. Process & change
  • Workflow redesign
  • New ways of working
  • Governance processes
  • Service redesign
  • Organisational change
4. Data & information
  • Data quality & remediation
  • Information management
  • Permissions & access
  • Knowledge management
  • Taxonomy & metadata
5. Assurance & risk
  • Security & privacy
  • Responsible AI
  • Testing & evaluation
  • Human review
  • Audit & compliance
6. Operations & continuous improvement
  • Monitoring & observability
  • Support & maintenance
  • Incident management
  • Benefit measurement
  • Iteration & enhancement

1. Discover & plan Identify needs, use cases and business value. Build the investment case.

2. Pilot & experiment Test, learn and validate value, risk and cost. Measure and refine.

3. Scale & adopt Roll out to teams and services. Build capability.

4. Operate & optimise Monitor usage, value and cost. Optimise models and processes.

5. Sustain & evolve Continually improve, adapt and reinvest for future value.

Why invest
  • Improve public services and citizen outcomes
  • Increase productivity and efficiency
  • Enhance decision making with better information
  • Reduce risk and strengthen compliance
  • Drive innovation and long-term value
Where value is realised
  • Faster, more consistent service delivery
  • Better experience for users and citizens
  • Lower cost to serve over time
  • Improved quality, accuracy and compliance
  • Stronger resilience and adaptability
Funding approaches

Central funding Departmental budgets Programme funds Shared services Outcome / value based funding Public-private partnerships

Measure what matters

Adoption rate Usage & engagement Quality & accuracy Cost per transaction Time saved Outcome & impact

Without full investment
  • Under-adoption and poor outcomes
  • Hidden costs and budget shocks
  • Security, compliance and quality risks
  • Duplication and technical debt
  • Lost trust and stalled programmes

Back to the contents · Suggest a correction (opens GitHub)

Part 6 · Fund and buyChapter 6.2 · 35 of 76

FinOps & models

A builder's design choice sets a budget holder's cost

AI cost is a portfolio, not a licence line. You are billed per token, in and out — roughly four characters or three-quarters of a word each; an A4 page is 600–800 tokens. Micro-costs stack when an agent runs on repeat. Report by service, product, agent, owner, model and cost driver; showback before chargeback; no expenditure classification without finance-led assessment.

Licences Credits Model calls Compute Storage Connectors Monitoring Evaluation Adoption & training Remediation Security Human review Support

How the two main vendors now charge, and where they process

Published list positions at 19 October 2026
Microsoft Copilot

Cost: a per-user licence ($30 per month list) covers Chat and Copilot in the Office apps. Cowork, Code, Autopilot and frontier models consume Copilot Credits at $0.01 each on pay-as-you-go. Agent security for custom agents needs Agent 365.

Residency: follows the tenant's Microsoft 365 geography. Web grounding, subprocessor models (Claude, Grok) and Code apps on Managed Runtime need separate checks.

Anthropic Claude

Cost: a seat per user, plus usage at token rates by model (Opus 5.5 lists at $4 input / $20 output per million tokens). Model routing, effort settings and prompt caching are the main levers.

Residency: set by deployment route. Anthropic-processed with a US-only option, or in-region through Bedrock, Google Cloud or Foundry. Connectors carry the content they fetch to the model.

See all 39 capabilities compared in the capability atlas →

Cost lab — the same job, five model classes

Illustrative list prices from the "Where your AI spend goes" explainer (2025), GBP per 1,000 tokens — verify current pricing before any decision. Non-English output can cost more: the same words take more tokens.

GPT-4o class (OpenAI) Balance of quality and tone £0.017per run £1.71per month

Claude Haiku class (Anthropic) Cost-efficient, high-volume work

£0.0043per run

£0.427per month

Claude Opus class (Anthropic) Deep, high-context threads

£0.026per run

£2.56per month

Mixtral class (Mistral) · cheapestUltra-low cost; needs prompt tuning

£0.0032per run

£0.320per month

Gemini Pro class (Google) On par with GPT-4o by usage tier

£0.017per run

£1.71per month

Design levers before model choice: reuse outputs, cache repeated answers, trim inputs, use plain logic before calling a model.

Four lenses of value — the estate case, measured and modelled

Reclaimed time
~£0.5M–£1.6M / yr at ~680 staff

A 10% findability gain on ~1.8 hrs/day spent searching. The big lever — and it can cover the licence bill 2.5–8× over ("earned allocation").

Storage cost ~£1.5k / yr Effectively £0 marginal under bundled M365. Symbolic.

Environmental
~9 TB ≈ 55–360 kg CO₂e / yr

Modest; aligns with Greening Government; curbs future growth.

Wellbeing
117 emails/day · ~23 min to refocus

Interrupted every ~2 minutes; 48% find work chaotic and fragmented. The human case.

Rules that hold: require total-cost and portability evidence before scaling any tool or agent · per-agent consumption caps and anomaly alerts · unit economics per production agent published within 12 months · value measured through the evidence chain: activity → output quality → workflow change → organisational outcome → public outcome.

Back to the contents · Suggest a correction (opens GitHub)

Part 6 · Fund and buyChapter 6.3 · 36 of 76

Commercial barriers · illustrative default scores

Barriers that slow adoption, increase cost and limit sustainability

Score each barrier by its impact and its likelihood in your context. Risk score is impact × likelihood; the band follows the published scale. The defaults are the sector-level assessment — change them and the bands move.

Critical

High
Medium
Low

4 barriers are critical and 8 high in this scoring. Address these before scale.

BARRIER IMPACT LIKELIHOOD I×L RISK TYPICAL CURRENT STATE

Licensing clarity Unclear boundaries between included features, credits, capacity and pay-as-you-go.

20
Critical
  • Moving boundaries
  • Preview versus general availability confusion

Cost & usage transparency Limited access to usage data, telemetry, forecasting and billing detail.

20
Critical
  • Trials paused
  • Inability to model total cost of ownership

Price predictability Usage-based pricing makes budgets volatile and hard to forecast.

16
Critical
  • Variable token costs
  • Capacity uncertainty

Supplier / platform lock-in High switching costs, proprietary connectors and formats.

16
Critical
  • Proprietary ecosystems
  • Data portability gaps

Telemetry & audit rights Lack of contractual rights to monitor, audit and export operational data.

12
High
  • Limited audit trails
  • Evidence gaps

Data ownership & IP Uncertainty over prompts, outputs, data use, training and generated code.

12
High
  • IP ambiguity
  • Data use concerns

Data residency & jurisdiction Data and model processing across jurisdictions may create compliance risk.

12
High
  • Cross-border processing
  • Sovereignty concerns

Service level & support Unclear SLAs, responsibilities and support across multiple products and models.

9
High
  • Shared responsibility
  • Support fragmentation

Interoperability & portability Limited open standards, export options and easy integration with other systems.

12
High
  • Custom connectors
  • Integration complexity

Commercial complexity Complex contracts, overlapping costs, multiple components and change over time.

9
High
  • Complex terms
  • Change without notice

Small body disadvantage Smaller organisations lack commercial leverage and specialist expertise.

12
High
  • Limited negotiation power
  • Resource constraints

Exit & transition risk Uncertain exit costs, data extraction, migration and parallel running.

12
High
  • Exit complexity
  • Transition cost risk

Bands are computed from the published impact × likelihood scale. Two rows on the source board carry a stated risk level that does not match its own band scale; the computed band is shown here.

Common examples
  • Trial paused due to lack of cost / usage data
  • Unexpected charges from token consumption
  • Unable to export data or prompts
  • Contract terms prevent model substitution
  • Data processed outside approved regions
  • No audit rights or limited visibility
  • SLA ambiguity across multiple services
  • Custom integrations create lock-in
Actions to reduce risk
  • Consolidate demand and negotiate as one government
  • Demand transparency, telemetry and audit rights
  • Use open standards and ensure portability
  • Define clear data ownership and IP terms
  • Require model and data location assurance
  • Build modular architectures and exit plans
  • Share commercial intelligence across the community
  • Support smaller bodies with frameworks and guidance

Build, buy, configure or share?

Match the approach to the need, risk, control, value and total cost. Decide wisely, start small, scale with confidence — and revisit the decision as needs, risk and technology evolve.

1. Define the needWhat problem are we solving?

2. Assess contextUsers, data, risk, regulation, criticality

3. Evaluate optionsUse the criteria to compare

4. Decide & planSelect approach, plan delivery

5. Review & adaptMeasure outcomes, learn and adjust

Use platform / pre-built

Moderate complexity / risk

Use existing platform features or pre-built agents with configuration.

Best for

Common needs, moderate risk, fast value

You get
  • Faster delivery
  • Lower cost
  • Built-in updates & support
Considerations
  • Limited customisation
  • Platform roadmap dependency
  • Data & permission alignment
Examples

Microsoft 365 Copilot, pre-built agents, Copilot Studio templates, AppSource solutions

Sample scoring matrix · 1 low to 5 high

Strategic value Risk & assurance Control & flexibility Total cost (TCO) Time to value Reusability & scale AVG Personal configuration 2 2 4 5 5 1 3.2 Use platform 3 3 3 4 4 3 3.3 Configure / extend 4 3 4 3 3 4 3.5 Build custom 5 4 5 2 2 3 3.5 Share / buy 4 4 3 4 4 5 4.0

Very high (strong) High (good) Medium (acceptable) Low (weak) Very low (poor)

Key questions to help decide
  • What is the impact if this goes wrong?
  • What data will it access or generate?
  • Do we need full control or can we share?
  • Is there an existing solution we can reuse?
  • What is the total cost over the lifecycle?
  • How quickly do we need value?
The continuous review loop

1. Monitor usage, value and risk 2. Evaluate outcomes and costs 3. Learn and improve approach 4. Adapt or reinvest

Common outcomes
  • Right solution for the need
  • Controlled risk and spend
  • Faster delivery and adoption
  • Reusable assets and knowledge
  • Better outcomes for citizens

From community intelligence to better contracts

A continuous loop that turns real-world experience across government into stronger commercial engagement, better terms and improved public value. Open once, use many times.

Inputs

Departments & public bodies Devolved governments Local government & combined authorities Health & care Police, justice & emergency services Education & skills Regulators & public agencies Communities of practice & forums Audits, reviews & external reports Financial & FinOps data

1
Community intelligence
  • Live problems
  • Use cases & pilots
  • Costs & usage data
  • Risks & barriers
  • Lessons learned
2
Consolidate & analyse
  • Thematic analysis
  • Prioritise issues
  • Identify patterns
  • Evidence strength
  • Impact assessment
3
Define requirements & questions
  • Functional needs
  • Non-functional needs
  • Data & telemetry
  • Interoperability
  • Commercial terms
4
Engage suppliers & market
  • Share requirements
  • Test assumptions
  • Request evidence
  • Validate options
  • Explore innovation
5
Refine terms & frameworks
  • Contract clauses
  • Pricing & models
  • SLAs & commitments
  • Exit & portability
  • Risk & liability
6
Implement & monitor
  • Deploy solutions
  • Monitor performance
  • Track usage & cost
  • Manage risks
  • Capture issues
7
Feedback & share
  • Share outcomes
  • Publish lessons
  • Update patterns
  • Support others
  • Improve guidance
8
Evolve & influence
  • Refine priorities
  • Influence roadmaps
  • Shape market
  • Strengthen policy
  • Drive innovation

Outputs

Better contracts & terms Cost transparency & predictability Stronger controls & governance Interoperability & portability Improved supplier performance Shared patterns & reusable assets Evidence-based policy & standards Better outcomes for citizens

Guiding principles

Evidence led Real experience, not assertion

Collaborative & inclusive Across departments, nations and sectors

Fair & proportionate Terms matched to risk and scale

Transparent & accountable Open about costs, terms and outcomes

Open once, use many times Publish so others do not repeat the work

Continuously improve The loop does not stop at signature

Enablers
  • Data & analytics
  • Skilled commercial & technical teams
  • Digital platforms & collaboration
  • Standards & guidance
  • Security & information assurance
Outcomes
  • Faster, safer adoption
  • Better value for money
  • Reduced risk & duplication
  • Stronger innovation & competition
  • Continuous improvement
  • Trusted, sustainable public services

Back to the contents · Suggest a correction (opens GitHub)

Part 6 · Fund and buyChapter 6.4 · 37 of 76

Dependency, lock-in & exit

The ecosystem behind an AI-enabled public service

AI-enabled services rely on a chain of suppliers, platforms, models, data sources and infrastructure. Visibility of these dependencies is what makes risk management, resilience, portability and value possible.

Public sector consumers

Central government departments
Devolved administrations
Local government
Public bodies & arm's-length bodies
Health & care
Education
Justice & policing
Regulators
Citizens & service users

AI-enabled public service / use case

Knowledge assistant Case summary Policy drafting Meeting intelligence Citizen support Code assistant Workflow automation Analytics & insight

Applications & platforms

Microsoft 365 Copilot Copilot Studio / Agent Builder Teams, SharePoint, Outlook Power Platform (connectors, apps, automate) Foundry / Azure AI services Third-party applications Custom applications

Models & AI capabilities

Microsoft models OpenAI models Anthropic models Google models Open source models Specialist / domain models Future / emerging models

Cloud & infrastructure

Microsoft Azure (compute, storage, networking) Azure regions & availability zones GPU / compute capacity providers AI services (API management, safety, content filters) Identity & access (Entra ID) Monitoring & observability Backup & disaster recovery

Data & information

SharePoint libraries Teams & collaboration content Line-of-business systems Open data & external sources Reference data & master data Vector stores & indexes Logs, telemetry & feedback

Third parties & suppliers across the ecosystem

Model providers Cloud provider Software vendors API / service providers Data providers Integration partners Support & operations Consulting & implementation

Key dependency categories

Platform dependencyReliance on specific platforms and ecosystems

Model dependencyReliance on model providers and availability

Data dependencyAccess to quality, timely and trusted data

Infrastructure dependencyCompute, storage, network and region reliance

Supplier & contractual dependencyThird-party services, SLAs, pricing and contractual terms

Operational dependencySkills, support, monitoring and continuity

Key questions to ask
  • What happens if a supplier or model is unavailable?
  • Can we switch models or platforms easily?
  • Where is our data stored and processed?
  • Do we have visibility of usage, cost and performance?
  • What contractual rights do we have — audit, data portability?
  • Are we building excessive lock-in or technical debt?
  • Do we have a clear exit and migration plan?
Managing dependencies
  • Map and document all dependencies
  • Assess risk and criticality
  • Ensure contractual protections
  • Design for portability
  • Monitor usage, cost and performance
  • Plan for resilience and exit
  • Review and adapt continuously
Outcomes

Better risk management Resilience & continuity Cost optimisation Informed supplier strategy Sustainable innovation Public value protection

Design for choice today. Exit with confidence tomorrow.

Avoid unintended dependency. Preserve portability across data, models, agents, integrations and knowledge, so you can adapt, switch or exit when you need to.

PortabilityKeep your data, prompts, models and agents portable

TransparencyKnow what you use, where it is, and the terms

InteroperabilityUse open standards and avoid proprietary traps

Plan aheadDesign exit and migration paths from day one

Control costsUnderstand true switching and transition costs

ContinuityMaintain service and compliance through change

Sources of lock-in
  • Proprietary platforms and ecosystems
  • Closed data formats and schemas
  • Proprietary connectors and APIs
  • Model dependency and exclusivity
  • Managed services and operations
  • Contractual restrictions and terms
  • Data residency and jurisdiction
  • Skills and tooling
The lock-in zone — where dependencies accumulate

Technology dependencies

  • Proprietary features and roadmaps
  • Custom integrations
  • Closed standards
  • Single-tenant architectures

Data & knowledge dependencies

  • Data locked in proprietary stores
  • Non-exportable metadata
  • Knowledge bases tied to platform
  • Prompt libraries not portable

Model & AI dependencies

  • Model exclusivity clauses
  • No right to substitute
  • Fine-tuned models not exportable
  • Training data use restrictions

Operational dependencies

  • Managed service entanglement
  • Monitoring and tool lock-in
  • Infrastructure coupling
  • Support and SLAs tied to stack

Commercial dependencies

  • Long contracts with penalties
  • Limited termination rights
  • IP and data use restrictions
  • Unilateral price change rights

Organisational dependencies

  • Specialist skills tied to vendor
  • Processes built around tools
  • Change resistance
  • Limited market knowledge
Exit & transition path — reduce risk, preserve value
1
Discover & inventory
  • Map all assets, services and data
  • Identify critical dependencies
  • Assess complexity and effort
2
Define target state
  • Choose target platform / model
  • Define data, security and compliance needs
  • Agree outcomes and constraints
3
Design migration strategy
  • Prioritise workloads
  • Phased or parallel approach
  • Define data and application migration
  • Plan cutover and rollback
4
Prepare & extract
  • Extract data, metadata and documents
  • Export prompts, agents and configurations
  • Validate completeness and quality
5
Build & test
  • Rebuild integrations
  • Adapt to target platform / model
  • Test functionality, performance and security
6
Migrate & cutover
  • Migrate data and configurations
  • Switch traffic and users
  • Monitor closely post-cutover
7
Optimise & decommission
  • Optimise costs and performance
  • Retire old assets safely
  • Capture lessons and update runbooks
Exit readiness check
Not assessed

Rate how prepared you are to exit. The path above is the work; this is the readiness.

What to preserve and make portable

Data Metadata & schemas Documents & content Prompts & instructions Agents & workflows Configurations & settings Integrations & APIs Access & permissions Audit logs & evidence Knowledge & context

De-risking strategies
  • Use open standards and portable formats
  • Negotiate data and IP rights up front
  • Retain audit and telemetry access
  • Stay multi-cloud and multi-model ready
  • Modular architecture and loose coupling
  • Regular dependency reviews
Red flags — high lock-in risk
  • No export or portability options
  • Data only accessible via proprietary APIs
  • High egress or extraction charges
  • Contract penalties for early exit or data retrieval
  • Complex custom integrations with no standards
  • Single model or provider dependency
  • Lack of roadmap visibility or support
Exit cost components
  • Data extraction and transformation
  • Rebuild / re-engineering
  • Dual running (parallel operations)
  • Testing and assurance
  • Training and change
  • Contract exit and penalties
  • Downtime / disruption risk
Exit success outcomes
  • Freedom to innovate and adapt
  • Lower total cost of ownership
  • Stronger resilience and continuity
  • Better negotiation power
  • Protected public data and trust
  • Sustainable, future-proof services

Back to the contents · Suggest a correction (opens GitHub)

Part 7 of 12 · 11 chapters

People and adoption

Part 7 · People and adoptionChapter 7.1 · 38 of 76

The team

Who needs to be in the room

The Service Manual's multidisciplinary team, extended for AI. Not every role is full-time, but every role is engaged from discovery, not summoned for sign-off. "Critical at" marks where each role's absence hurts most.

Set the direction

Accountable and delivery leadership

Senior Responsible Owner

Accountability, benefits, risk appetite. The name on the assurance forms.

Critical at · Discover → Operate

Product manager

Problem framing, value, prioritisation of the backlog.

Critical at · Discover · Design

Delivery manager

Cadence, dependencies, unblocking the team.

Critical at · Build · Deploy

Programme / portfolio lead

Fit with the wider portfolio; stops duplicate AI pilots.

Critical at · Discover · Deploy

Build the capability

Data and AI engineering

Data scientist

Modelling choices and honest evaluation design.

Critical at · Design · Build

AI / ML engineer

Pipelines, integration, MLOps, guardrails in code.

Critical at · Build · Operate

Data engineer

Data quality, provenance, pipelines that last.

Critical at · Design · Build

Technical architect

System design, interoperability, exit routes.

Critical at · Design

Assure and protect

Risk, rights and control

Cyber security specialist

Threat modelling and NCSC-aligned controls.

Critical at · Design · Deploy

DPO / information governance

DPIA, lawful basis, records of processing.

Critical at · Discover · Design

Legal adviser

Public law, IP and copyright, contracts.

Critical at · Design · Deploy

Ethics adviser

Bias, participation, societal impact.

Critical at · Discover → Operate

Ground it in reality

Users, domain and money

User researcher

Real user needs, and real harms, observed rather than assumed.

Critical at · Discover · Operate

Service designer

The end-to-end journey, including the failure paths.

Critical at · Design

Domain expert / operational staff

Casework truth, edge cases, what the data actually means.

Critical at · Discover → Operate

Commercial specialist

Procurement routes, vendor terms, ethical parity for third parties.

Critical at · Discover · Deploy

Skills pipeline. Recruit and upskill through the Government Digital and Data profession's capability framework, Civil Service AI training and the cross-government AI community. Playbook principle 9 makes this the deploying organisation's responsibility, not the vendor's.

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.2 · 39 of 76

Multidisciplinary AI service team

Twelve disciplines around one AI service

Collaborating to design, build, govern and operate trustworthy AI services that deliver public value — through design and plan, build and enable, deploy and operate, measure and improve. The service at the centre is defined by user needs, public value, trust, safety and outcomes.

Product management

  • Strategy & roadmap
  • Prioritisation & value
  • Stakeholder alignment
Service design
  • End-to-end journeys
  • Inclusive & accessible
  • Experience design
Solution architecture
  • Architecture & patterns
  • Integration & interoperability
  • Scalability & resilience
Data & AI
  • Data strategy & governance
  • Data quality & lineage
  • Model design & evaluation
Security
  • Secure by design
  • Threat modelling
  • Security assurance
Information governance
  • Information management
  • Records & retention
  • Data classification
Legal & policy
  • Lawful & ethical use
  • Policy compliance
  • Contracts & liabilities
Commercial
  • Procurement strategy
  • Supplier management
  • Value for money
Operations
  • Run & support
  • Monitoring & reliability
  • Incident management
Learning & development
  • Capability & skills
  • Training & guidance
  • Adoption & culture
Accessibility & inclusion
  • WCAG & inclusive design
  • Equality impact
  • Usability for all
User research
  • Understand users
  • Research & insights
  • Usability & validation

Team principles

User-centredDesign with and for users at every stage.

InclusiveAccessible, equitable and easy to use.

Trust & safetyBuild security, privacy and ethics in from the start.

CollaborativeWork together, share ideas and challenge constructively.

Agile & iterativeLearn quickly, adapt and improve continuously.

Outcome focusedMeasure what matters and deliver public value.

Key inputs

  • User needs & insights
  • Policies, strategy & standards
  • Data & knowledge assets
  • Technology & platforms
  • Stakeholders & partners

Key outputs

  • Trusted AI services that work for users
  • Better outcomes & public value
  • Evidence & insights for decisions
  • Assured, compliant & well governed
  • Skilled teams & continuous learning

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.3 · 40 of 76

Roles & pathways

Learn. Apply. Progress.

Role-specific learning journeys to build skills, confidence and impact at every stage of adoption. Pick the role you are supporting; the pathway shows what capability looks like at each of four levels.

YOUR JOURNEY

1. Assess current skills

2. Choose your pathway
3. Learn & practise with support
4. Apply in real work
5. Measure impact & improve
1 · Aware

Understand AI basics and opportunities

AI fundamentals
Opportunities & risks
Strategic context
Responsible AI
2 · Practitioner

Apply AI tools and techniques with guidance

Use cases & value
Prioritisation
Governance basics
Supporting teams
3 · Advanced

Design, adapt and lead AI solutions

AI strategy alignment
Portfolio oversight
Investment decisions
Risk appetite
4 · Expert

Shape strategy, standards and innovation

AI vision & innovation
Ecosystem leadership
Policy influence
Ethical leadership

What supports progress at every level

Learning & developmentRole-based curricula, microlearning, labs, communities of practice

Tools & platformsApproved tools, secure sandboxes, reusable components

Guidance & standardsPolicies, playbooks, architectures, guardrails and templates

Measuring progressSkills assessment, certification, impact and value metrics

Recognition & careerCareer pathways, communities, badges and incentives

Principles behind the pathways

Role relevant Right skills for the job

Practical Learn by doing in real work

Progressive Build from foundations to mastery

Trusted Safe, secure and responsible AI use

Measurable Evidence of competence and impact

Inclusive Accessible and supportive for all

From task execution to higher-value work

AI augments people, it does not replace them. Roles evolve to focus on judgement, creativity, oversight and continuous improvement.

ROLE TRADITIONAL FOCUS AI ENABLES AI-ENABLED FUTURE FOCUS IMPACT

Policy adviser

  • Research and information gathering
  • Drafting options and text
  • Compiling background papers

AI summarises evidence, suggests options, drafts content, finds patterns

Strategic analysis and judgement
Evaluating trade-offs and impacts
Engaging stakeholders and deciding

Better policy quality and impact

Contact centre adviser

  • Answering routine queries
  • Searching knowledge bases
  • Logging and updating cases

AI handles routine queries, retrieves answers, suggests next best action

Complex case resolution
Empathy and relationship building
Handling exceptions and escalation

Improved citizen experience

Software engineer

  • Writing and refactoring code
  • Testing and fixing defects
  • Reviewing changes

AI generates code, finds bugs, recommends tests, documents changes

Architecture and design decisions
Code review and security assurance
Solving novel and complex problems

Faster delivery, higher quality

Data analyst

  • Data extraction and preparation
  • Running standard reports
  • Creating charts and dashboards

AI prepares data, generates insights, builds visuals, detects anomalies

Deeper analysis and insight
Explaining drivers and implications
Advising on data strategy

Stronger insight and decisions

HR advisor

  • Processing requests
  • Answering policy questions
  • Updating records

AI answers FAQs, guides processes, drafts documents, checks compliance

People strategy and coaching
Complex case management
Culture and inclusion leadership

Better employee experience

Risk manager

  • Collecting risk data
  • Maintaining risk registers
  • Producing standard reports

AI identifies risks, analyses scenarios, monitors controls, suggests mitigations

Interpreting risk in context
Prioritising actions and trade-offs
Overseeing governance and assurance

Stronger risk management

The shift
From doing tasks…
…to orchestrating with AI
…to focusing on judgement, creativity and impact
Delivering better outcomes for citizens and society
What people need
  • AI literacy and critical thinking
  • Change mindset and adaptability
  • Collaboration and communication
  • Ethics and responsible use

Success depends on people, process redesign, culture, leadership and trust — supported by responsible AI and strong governance.

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.4 · 41 of 76

AI workforce capability academy

Building the skills and confidence to design, develop, govern and operate trusted AI services

Five learning pathways across four stages. Build confidence. Apply skills. Drive impact. Lead change.

FoundationUnderstand the basics BuildDevelop practical skills AdvanceApply and lead LeadExpertise and influence

Leaders & decision makers

Set direction and provide oversight

Foundation AI essentials for leaders Opportunities, limitations · Risks and public value · Responsible AI principles

Build AI governance & accountability Oversight responsibilities · Risk appetite and decision rights · Human control and assurance

Advance Strategic AI leadership Portfolio management · Measuring outcomes and impact · Ethics, bias and inclusion

Lead Shaping the AI agenda Leading culture change · Stakeholder engagement · Policy influence and advocacy

Product & service teams

Design and deliver AI-enabled services

Foundation AI for product & service teams User needs and problem framing · AI use cases and value · Ethical and inclusive design

Build Designing AI-enabled services Journey mapping with AI · Prototyping and evaluation · Managing suppliers

Advance AI product management Prioritisation and roadmapping · Data-informed outcomes · Scaling AI responsibly

Lead Leading AI transformation Cross-team leadership · Change management · Driving adoption and value

Technical & data teams

Build and run secure, reliable AI agents

Foundation AI & data fundamentals Data types and quality · Intro to ML, LLMs and agents · Privacy and security basics

Build Building AI agents Prompt engineering · RAG and tool integration · Testing and evaluation

Advance Advanced engineering Agent orchestration · Model evaluation & guardrails · Red teaming and robustness

Lead Technical excellence Architecture leadership · Innovation and emerging tech · Open standards and reuse

Governance & assurance teams

Assure, audit and manage risk

Foundation AI governance basics Policies, standards and roles · Risk types and controls · Transparency and ATRS

Build Assessing and controlling risk DPIA for AI systems · Bias, fairness and inclusion · Security threat modelling

Advance AI assurance & audit Assurance frameworks · Monitoring and evidence · Incident and issue management

Lead Assurance leadership Independent challenge · Audit strategy and reporting · Regulatory horizon scanning

Operational & support teams

Operate, monitor and support users

Foundation Working with AI tools Everyday AI use · Human oversight basics · Data handling and security

Build Operating AI services Monitoring and logging · Incident reporting · User support and feedback

Advance Optimising operations Performance analysis · Continuous improvement · Knowledge sharing

Lead Operational excellence Service reliability leadership · Automation and productivity · Building a learning culture

Core capabilities for all roles

Responsible AIEthics, fairness, transparency

Security & privacySecure by design, data protection

Human oversightAccountability, challenge, escalation

Data literacyQuality, lineage, classification

Accessibility & inclusionDesign for everyone, equitable outcomes

Continuous learningIterate, adapt and improve

Learning methods

Instructor-led workshops E-learning modules Hands-on labs & sandboxes Communities of practice Certification & badging Playbooks & job aids

Capability progression

Aware Practitioner Expert Leader

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.5 · 42 of 76

Capability & maturity · illustrative frameworks

Maturity is a journey, not a destination

Seven levels, from ad-hoc use to adaptive capability. Choose a level to see what it looks like across the dimensions and the twelve capability domains — then record the evidence you could actually show.

Level 3 · Governed

Next: 3 → 4 — Integrate and embed

Defined governance and controls. Repeatable patterns, registered use cases and measured pilots.

Leadership Accountable ownersGovernance Controls, risk tiers, assurance gatesPeople & skills Role-based skills
Data & knowledge Catalogued, governed accessTechnology Reference architecture, model choiceValue & outcomes Measured value from pilots

Evidence over opinion

Record the evidence you can actually show for each domain.

Strong Independently validated

Moderate Documented internal evidence

Emerging In pilot or partially proven

Weak Anecdotal or not evidenced

Not assessed No evidence available

Strategy & leadership
Not recorded

AI strategy aligned to business objectives and risk appetite

People & skills
Not recorded

Role-based skills framework and capability plans

Use cases & value
Not recorded

Use case portfolio managed with benefits tracking

Technology & architecture
Not recorded

Target architecture, standards and integration patterns

Data & information
Not recorded

Governed data assets, metadata and lineage

Security & privacy
Not recorded

Security-by-design; identity, access and monitoring in place

Responsible AI & ethics
Not recorded

Responsible AI framework and assurance process

Commercial & suppliers
Not recorded

Managed suppliers, contracts and risk assessments

Finance & FinOps
Not recorded

Cost visibility, allocation and showback

Operations & service management
Not recorded

Defined operating model and SLAs

Measurement & outcomes
Not recorded

Benefits tracking; outcomes measured

Community & knowledge
Not recorded

Communities of practice and curated knowledge

How to use this
  • 1 Assess current evidence against each domain
  • 2 Identify priority gaps — emerging and weak ratings
  • 3 Take actions to build evidence and capability
  • 4 Review and update regularly
Evidence that counts

Policies & strategies Architectures & standards Training records Use case business cases KPIs & dashboards Audit & assurance reports Lessons learned & patterns

What good looks like
  • Clear strategy and governance
  • Skilled and empowered people
  • Trusted data and knowledge
  • Secure and responsible by design
  • Smart commercial and supplier management
  • Efficient and sustainable spend
  • Measurable outcomes and public value
  • Strong communities and shared learning

Where the capability gaps are, by role

Current against required capability, scored 5 (critical gap) to 1 (strong, sustain). Illustrative sector-level assessment — replace with your own to prioritise investment. Filter to one role group to read a single column.

CAPABILITY DOMAINExecutives & leadersBusiness & service ownersProduct & delivery managersData & info professionalsEngineers & technical teamsSecurity & risk professionalsCommercial & procurementFinance & FinOpsFrontline staff & practitioners
Strategy & leadership Vision, governance, value244433344
People & change Adoption, culture, engagement333344334
AI & data literacy Understanding AI, data and information343233435
Technology & engineering Tools, platforms, integration333323444
Data & information Quality, access, governance444233334
Security, privacy & risk Threats, controls, compliance333432334
Responsible AI & ethics Fairness, transparency, human oversight334333334
Commercial & procurement Contracts, suppliers, value334333244
Finance & FinOps Cost management, allocation, value343333323
Operations & service management Monitoring, support, continuity343333333
Measurement & outcomes KPIs, impact, benefits — not scored on the source board—————————

Critical gap High priority

Major gap Act soon

Moderate gap Plan to address

Minor gap Monitor

Strong / no gap Sustain

Top priority gaps

AI & data literacy for frontline staff

Data & information for executives, business owners and product managers

  • People & change for technical and security roles
  • Commercial & procurement for product managers and finance
  • Finance & FinOps for product managers
Key insights
  • Frontline staff need stronger AI literacy and confidence
  • Data and information capabilities are a cross-role enabler
  • Product managers and finance need deeper commercial and FinOps skills

Technical teams are relatively strong but need continued investment in security by design

How to use the heatmap
  • 1 Assess current capability levels by role
  • 2 Compare to required capability levels
  • 3 Prioritise high-impact gaps
  • 4 Invest in learning, support and resources
  • 5 Measure progress and outcomes
Actions to close gaps

Role-based learning pathways Communities of practice and peer learning Targeted coaching and enablement Tools, templates and reusable assets Time, budget and leadership support

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.6 · 43 of 76

Inclusive & accessible adoption

Different people. Different needs. Shared success.

AI tools should empower everyone by offering choice, clarity and support. The spectrum below runs from empowered and confident to apprehensive — and inclusion is not one size for all: listening, choice and support help everyone succeed with AI.

Enthusiastic & confident

Sees AI as a productivity booster and enjoys exploring new tools.

May typically
  • Experiment and customise
  • Share tips and best practice
  • Integrate AI into daily work
Key needs
Advanced features
Experimentation space
Recognition & influence
Ways to contribute back
Cautious but willing

Interested and willing to try, but wants reassurance and guidance.

May typically
  • Use AI for simple, low-risk tasks
  • Seek examples and best practice
  • Prefers structured learning
Key needs
Clear guidance & examples
Safe practice environments
Step-by-step learning
Trusted peer support
Needs support & clarity

Wants to benefit from AI but needs help understanding how and when.

May typically
  • Uses guided features
  • Needs help in context
  • Prefers practical, relevant training
Key needs
Plain language & relevance
Role-based training
Hand-holding & coaching
Time to learn & practise
Anxious & overwhelmed

Feels unsure or overwhelmed by change, complexity or risk.

May typically
  • Avoids or minimises use
  • Worried about errors or job impact
  • Prefers human interaction
Key needs
Reassurance & transparency
Simple, limited choices
Human support always available
Safe, low-pressure options
Resistant / disengaged

Does not see value or may distrust AI or the organisation's intent.

May typically
  • Declines to use
  • Questions fairness or reliability
  • Prefers current ways of working
Key needs
Listen & understand concerns
Involve in decisions
Address trust & fairness
Show tangible benefits
Design for inclusion

Choice & controlLet users choose how AI helps, when and in what way

Clear & simpleUse plain language, short explanations and examples

Flexible interactionSupport different workflows, speeds and preferences

Neuroinclusive designReduce cognitive load, provide structure, minimise distractions

Accessible by defaultWCAG compliant, multimodal, and assistive-tech friendly

Safe & trustedProtect data, explain limits, and make AI behaviour predictable

Human supportEasy access to help, feedback channels and champions

Enablers & supports

Leadership sponsorship Digital & AI champions Inclusive communication Role-based learning paths Protected learning time Local support networks Feedback & continuous listening Measure & adapt for all users

Remember

Inclusion is not one size fits all. Listening, choice and support help everyone succeed with AI — and inclusive adoption empowers people, strengthens trust and delivers better public services.

The cross-government community constellation

Working together across communities to share knowledge, avoid duplication and build trusted, reusable AI and Copilot capabilities. The weekly cross-government call is the open forum at the centre.

Practice and engineering

Hands-on development, experimentation and engineering excellence.

AI Engineering Lab Data Science & AI Community Low Code Community Microsoft 365 Technical Community Prototyping and Sandboxes

Adoption and digital workplace

User enablement, organisational change and the digital workplace.

Digital Workplace Community Change & Adoption Network AI Champions Network Learning & Capability Network Comms & Engagement Network

Data management communities

Data governance, catalogues and information management excellence.

Data Management Network Data Catalogue Community Information Governance Network Metadata & Standards Community Knowledge Management Community

Ethics, safety and assurance

Ethics, AI governance and assurance across the public sector.

Cross-Government Data & AI Ethics Community AI Assurance Network Security & Privacy Community Accessibility & Inclusion Network

Governance and policy network

Policy, standards, guidance and compliance.

Government Digital Service (GDS) DSIT AI Policy & Standards Government Analysis Function Identity & Access Community Service Design Network

Commercial, finance and suppliers

Commercial strategy, vendor engagement and cost insight.

Commercial Community FinOps Community Supplier Engagement Community Contract & Licensing Community Procurement Policy Community

Cyber and risk network

Cyber security, risk management and resilience.

Cyber Security Network Risk & Assurance Community Resilience Community

Digital workplace communities

Adoption, change, champions and enablement.

Copilot Weekly XGov Call Community of practice sessions Show and tell forums

Cross-sector and place-based connectors

Central government departments & ALBs Devolved administrations Local government & combined authorities Health & care network Education network Policing & justice network International & supranational links

Why the constellation matters

Share once, use many timesPublish once so others do not repeat the work

Avoid duplication and reworkFind what exists before building it again

Consistent standards, safer adoptionCommon expectations across organisations

Better value and lower total costAggregate demand and reuse assets

Faster innovation with confidenceLearn from those who have already tried

Better outcomes for citizensThe point of all of it

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.7 · 44 of 76

Journeys · adoption operating model

Adoption is a journey we take together

A system of people, processes and support, not a rollout. Leadership sets direction. Business owns outcomes. Communities enable change. Data and evidence drive continuous improvement.

People first

Trust & responsibility
Outcome focused
Iterate & improve
Collaborate & share
Evidence & value

Inputs

Strategy & priorities User needs & frontline insights Data, content & tools Policy & regulatory requirements Evidence & external intelligence Budget & investment

1
Lead & sponsor
  • Executive sponsorship and vision
  • Align to strategy and public value
  • Set priorities and risk appetite
  • Allocate investment and authority
  • Communicate the why
2
Engage & align
  • Business ownership and champions
  • Stakeholder mapping and engagement
  • Co-design with users
  • Change impact assessment
  • Inclusive and accessible by design
3
Prepare & enable
  • Skills, training and role-based learning
  • Guidance, guardrails and policies
  • Tools, environments and data readiness
  • Community of practice and peer support
  • Address accessibility and neurodiversity
4
Adopt & use
  • Start small, iterate and learn
  • Embed in workflows and processes
  • Measure early value and experience
  • Support and coaching at point of need
  • Promote quick wins and success stories
5
Measure & improve
  • Track usage, quality and outcomes
  • Measure value, cost and productivity
  • Gather feedback and identify issues
  • Improve models, processes and data
  • Report transparently and share learnings
6
Scale & sustain
  • Scale proven use cases and patterns
  • Standardise where appropriate
  • Manage risk, compliance and assurance
  • Optimise cost and resource use
  • Sustain, adapt and reinvest for impact

Outcomes

Better public services Improved user experience Lower cost, higher productivity Stronger compliance and trust Sustainable and adaptable Measurable public value

Enabling foundations

Data & information foundations
Security, privacy & safety
Responsible AI, ethics & inclusion
Technology & platforms
Policies, standards & governance
Commercial & FinOps
Communities & knowledge sharing
Operations & support

Measurement & feedback loop

1. Observe usage, quality, cost, risk 2. Analyse results, feedback, insights 3. Learn what works, what doesn't 4. Decide prioritise actions and changes 5. Act improve, enable, remove barriers 6. Repeat continuous improvement

Key roles

Leadership Set direction, sponsor change and unblock

Business owner Define outcomes, prioritise use cases and realise value

Product / service manager Manage lifecycle, measure value and iterate

Champion network Advocate, share learning and support peers

Data & information professional Ensure quality, access, lineage and governance

Security & privacy Manage risk, protect data and users

Commercial & FinOps Manage contracts, licensing, cost and value

Change & adoption team Drive engagement, training and communications

Technical team Build, integrate and operate solutions

Frontline users Use, feed back and help shape improvements

The person-level journey

From training to organisational capability

Six stages from awareness to community contribution — a progression, not a course. The continuous improvement loop runs across all six.

Stage 1 of 6 All staff

Awareness
Understand the opportunity
Goal

Build awareness of AI basics and potential in public services.

Key activities
  • Introductory sessions
  • AI fundamentals
  • Real-world examples
  • Responsible AI awareness
Outcome
Informed and curious workforce
Support model
  • Central resources and guidance
  • Local support and champions
  • Communities of practice
  • Expert clinics and labs
  • Feedback and improvement
Learning modes
  • Self-paced learning
  • Instructor-led training
  • On-the-job practice
  • Peer learning and sharing
  • Microlearning and nudges
Measuring progress
  • Adoption rates
  • Capability assessments
  • Business outcomes
  • User satisfaction
  • Value realised
Enablers throughout the journey

Leadership and sponsorship Clear guidance and guardrails Accessible tools and environments Role-based learning and support Time to learn and experiment Recognition and career pathways Feedback and user research Measurement and value realisation

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.8 · 45 of 76

Copilot programme · agent rollout

Why we decided not to just switch on first-party and Agent Builder agents

The opportunity is significant, but ownership, risk and consistency matter.

Agents can influence real work

More than simple prompts, built well, they can support processes, provide advice, surface information and shape decisions.

Creation creates responsibility

Every agent needs a clear purpose, an accountable owner and active lifecycle management. Risk management is critical.

Scale can amplify inconsistency or provide opportunity

Through a culture of safe experimentation, coupled with close monitoring, we will find opportunities to deliver the best agents, while avoiding “agent-aggedon”.

Success is confident adoption and organisational maturity — not simply the speed of feature enablement.

Finding the sweet spot — a “test & learn” approach

Innovate confidently. Manage risk. Deliver value. Together.

Experimentation at the core
Start small
Test & learn
Iterate quickly
Learn & adapt

Delivering

Outcomes that matter
Business value
User adoption
Continuous improvement
Sustainable scale
Start small. Learn safely. Scale with confidence.

Public Beta and measured communications ramp up gives the organisation space to learn about the technology, ownership and responsibility before broader scale.

Copilot Agent Builder is one of many business tools

Assurance requirements vary by risk and operational impact

Business areas tailor guidance to their environment

Governance that enables, not blocks

Proportionate guardrails create confidence to experiment.

Proactive, not reactive

Put standards and guidance in place early.

Central guardrails, local accountability

Set common expectations; keep responsibility with agent owners.

Proportionate and risk-based

Apply controls according to use case and data sensitivity.

Test & learn

Evolve the model using real scenarios and evidence.

What good looks like
  • Agents are visible, owned and managed through their lifecycle.
  • Responsible-use standards are embedded into governance and guidance.
  • Oversight covers usage, value and risk signals.
  • High-value use cases can progress without unnecessary friction.
  • Experimentation remains welcome within clear guardrails.
Governance is the licence to innovate confidently.

Support around every colleague

Technology enablement is paired with readiness, guidance and local ownership.

Clear guidance

What colleagues can use, when and how to work within the guardrails.

Communications

Initial soft launch, then expand messages around use cases and value. Align to OBT.

Peer support

Digital Champions, Sponsors and business representatives help locally.

Learning

Options from “getting started” through to building agents, focused on real use cases.

Adoption support

Tailored help to embed useful agents into team workflows. May extend to Tiger Teams.

Programme sets the foundation for safe, consistent and responsible use
Business areas adapts to local needs and drives adoption in their context

Measure what matters

Reporting is not the outcome. It is the feedback loop that helps us decide what to support, change, scale or retire.

Builders Who is creating agents?

Ownership Are agents visible and actively owned?

Usage Which agents are being used?

Enablement Are learning and communications changing behaviour?

Value What outcomes and benefits are emerging?

Challenges we are facing into
  • Reporting will mature
  • Identify duplication — spot the best agents
  • Alert when agents breach a usage or share threshold

Business takes ownership of their use cases and determines the best tool for the job

Local resources to take action on the measures

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.9 · 46 of 76

People and adoption · community

From isolated questions to collective capability

How a cross-government community matures: the engagement arc, the flywheel that turns experience into shared assets, and the honest exchange of what members give and get. Capability without community repeats mistakes; community without operationalisation produces discussion.

The engagement arc

Information access & orientation
Establishing the forum and enabling participation
  • Join meetings
  • Access chat and resources
  • Understand purpose
  • Share initial concerns
  • Learn community norms
Exploration & safe experimentation
Understanding tools and testing possibilities
  • Product changes & updates
  • Prompting and capabilities
  • Early pilots and proofs
  • Controls and guardrails
  • Licensing questions
Use cases & practical learning
Sharing examples and learning what works
  • Show & tell sessions
  • Success stories
  • Failures and lessons
  • Templates and patterns
  • Emerging use cases
Governance & assurance emerge
Defining guardrails and ways of working
  • Ownership and accountability
  • Data, security & privacy
  • Information management
  • Human oversight
  • Assurance approaches
Adoption & scaling together
Enabling confident organisational adoption
  • Adoption strategies
  • Change & comms
  • Training & champions
  • FinOps & cost visibility
  • Scaled rollouts
Capability & continuous improvement
Embedding capability and driving improvement
  • Measuring outcomes
  • Community feedback loop
  • Reusable assets
  • Influencing suppliers
  • Evolving standards

How we engage

Inclusive participation
Open questions & discussion
Share knowledge & experience
Challenge & constructively critique
Capture decisions & actions
Feedback & iterate
Enablers
  • Safe space
  • Regular cadence
  • Diverse perspectives
  • Community sponsorship
  • Accessible resources
Outcomes

Greater clarity & confidence Less ambiguity about products, controls and next steps.

Reduced risk & duplication Shared lessons prevent repeated mistakes.

Reusable knowledge Patterns, templates and guidance that can be adapted.

Faster & safer implementation Less rework, clearer pathways, better preparation.

Stronger supplier engagement Consolidated questions and influence based on community insight.

Value created
Better public services
Better use of public money
Increased trust & assurance
Continuous improvement of AI capability

Continuous learning loop Experience Insight Guidance Action Impact Feedback Improved practice

The collective learning flywheel

The more we share, the more we all improve.

01
Bring questions & challenges

Surface real needs and uncertainties

02
Share evidence & experience

Contribute what you have seen and tried

03
Compare, challenge & analyse

Learn together and test ideas

04
Create insights & patterns

Turn learning into shared assets

05
Apply, adapt & implement

Put learning into practice locally

06
Report outcomes & lessons

Share results and what is next

Inputs from the community
  • Real problems & open questions
  • Use cases & demonstrations
  • Challenges, risks & failures
  • Policies, approaches & processes
  • Cost, licensing & FinOps data
  • Security, privacy & assurance issues
  • Ideas, suggestions & future needs
Outcomes for all
  • Practical guidance & templates
  • Stronger guardrails & assurance
  • Reusable patterns & components
  • Skills, confidence & capability
  • Smarter investment & cost control
  • Influence with suppliers & partners
  • Better services & public value
Enablers of the flywheel
  • Inclusive & welcoming culture
  • Safe spaces for openness
  • Accessible knowledge & tools
  • Clear communication & feedback
  • Recognition of contribution

The value exchange

What members contribute, what the community turns it into, and what everyone gains.

What we contribute

Our questions & challenges Share real problems, uncertainties and barriers

Use cases & demos Show what we are building, testing and learning

Successes & lessons Share what works, what did not and why

Policies & approaches Contribute guidance, guardrails and governance ideas

Tools & templates Share artefacts, patterns, checklists and assets

Commercial & cost insight Raise licensing, FinOps and procurement experiences

Security, privacy & assurance Share risk findings, controls and assurance practice

Our time & expertise Offer help, review, connect others and support each other

Open discussions & peer challenge Honest conversations that surface real issues

Shared evidence & patterns Collective intelligence from across government

Expert input & supplier access Access to Microsoft, experts and the wider network

Connections & collaboration Find partners, avoid duplication and work together

Influence & impact Consolidate our voice to shape products, policy and standards

Capability uplift & confidence Improve skills, maturity and organisational readiness

Early awareness & reducing risk Spot changes, risks and opportunities earlier

Reusable outputs & guidance Turn learning into practical guidance and tools

What we gain

Practical answers Faster solutions to real problems

Time saved Learn from others' experience, avoid repeating mistakes

Safer adoption Better decisions with shared assurance and guardrails

Better quality outcomes Access to proven patterns and worked examples

Cost confidence Understand licensing, usage and FinOps more clearly

Stronger relationships Collaborate across departments, sectors and professions

Measurable progress Track maturity, impact and value together

Greater public value Better services, better use of public money, more trust

Shared principle: be open, be respectful, challenge constructively and focus on learning together to deliver safe, inclusive and effective AI in the public sector.

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.10 · 47 of 76

Operating model & programme

One operating model. Everyone aboard.

One AI capability operating model — not separate Copilot, agent, data, assurance and FinOps programmes. Cognitive load down, recognition up, DEIA by default rather than afterthought.

Four defaults, baked in

Privacy by default

Every product, every build. DPIA in, not post.

Existing

Secure by default

Open, auditable, no vendor lock-in.

Existing

DEIA by default

Inclusion, equity, access — in every spec.

Adding now

Recognition by default

Named people, visible work, earned respect.

Adding now

Reducing cognitive load

Less red tape

Decision logs replace approval chains

AI drafts briefings; you review

Single source of truth per project

Open by default, restrict by exception

Cross-pollination

Weekly cross-team show-and-tells

Shared prompt and tool library

SME pairing across directorates

Secondments and shadow programmes

Leadership engagement

Monthly programme board updates

Outcome dashboards, not status reports

Senior sponsors own a use-case

Ministers briefed on citizen impact

DEIA by default — who this serves, and the checklist

Every product must actively serve those experiencing the greatest adversity: economic hardship, disabled and neurodivergent users, underrepresented communities, vulnerable households and carers, digital exclusion and low literacy, and everyone algorithmic unfairness would otherwise reach first.

Who is most harmed if this fails or excludes?

Does the design assume internet access or literacy?

Has it been tested with people from affected groups?

Are legal obligations (Equality Act, PSED) documented?

Is there a non-digital fallback route?

Does the data we collect reflect the full population?

Are there language and accessibility alternatives?

How will we measure equitable outcomes, not just usage?

Bias will be reproduced unless actively constrained

A model trained on historical outcomes learns the discrimination in the record, not the potential of a well-supported person. Four pre-specified bias domains to audit — and abstention ("insufficient data — human decision only") is always acceptable, and must be possible.

Allocation

Systematically fewer or lesser recommendations for protected groups with equivalent profiles.

Calibration

Predictions accurate against a biased historical record, wrong about supported potential.

Representation gap

Too few cases for a group; the model fills gaps with majority priors.

Feedback loop

Biased outputs produce worse outcomes that retrain the model — break the loop by design.

Training-data standards: exclude invalidated-treatment-era and sanctioned-employer records and proxy features; require minimum cases per group, representation audits, labelled synthetic augmentation for regularisation only, and temporal weighting toward modern practice. Lived-experience panels review real outputs; community bodies sign off the test criteria; explanations reference functional needs, never the diagnosis.

The capability programme — four phases

Phase 0 · FoundationWeeks 1–4

Starter capability, safely

Private, data-safe assistant; no lock-in; GDPR by design

Data protection baseline

DPIA template, handling principles, red lines

Use-case registry

Crowdsource the top jobs-to-be-done; impact × feasibility

Phase 1 · PilotMonths 2–4

Learning pathway

Prompting fundamentals, hands-on sprints, peer cohorts

Open-tool pilots

Open-weight models on gov cloud; avoid lock-in from day one

Impact measurement

Hours saved, quality uplift, citizen outcomes — not vanity metrics

Phase 2 · ScaleMonths 5–9

Shared platform

Modular, containerised infrastructure across directorates

Champions network

One champion per team; shared prompt library; show-and-tells

Legacy reduction

AI-assisted modernisation, documentation, automation

Phase 3 · EmbedMonths 10–18

Public trust framework

Transparent registers, explainability standards, citizen comms

Recognition & visibility

Awards, case-study library, cross-government showcases

Continuous learning

Evergreen curriculum, AI-native onboarding, horizon scans

100%

team with baseline AI literacy

≥10

measurable use-cases live

0

proprietary lock-in

DPIA

embedded in every build

Named

civil servants celebrated

The sequence — three, six, twelve, twenty-four months

3 months

Establish preconditions
  • Reachability audit and permissions remediation
  • Publish agent-register schema and intake form
  • Consumption dashboards and per-agent caps
  • Agree retention and FOI position

6 months

Govern
  • Operate the register; classify every agent by tier
  • Apply assurance tiers; define the orphaned-agent process
  • Secure exit, portability and audit rights
  • Publish an evaluation method standard

12 months

Integrate
  • Fund adoption as a standing operating line
  • Publish unit economics per production agent
  • Unify information and agent governance
  • Publish one outcome measure, not an activity

24 months

Federate
  • Share registered agent patterns across bodies
  • Aggregate demand for small public bodies
  • Prove exit is real via a live substitution
  • Publish a community value evaluation

Sequence matters more than speed: permissions before scale, register before governance claims, cost instrumentation before volume. Recognition runs throughout — named people, visible work, credited contributions, minister-level visibility on citizen wins.

Back to the contents · Suggest a correction (opens GitHub)

Part 7 · People and adoptionChapter 7.11 · 48 of 76

Responsible AI · people and adoption

People at the centre: the adoption, capability and change model

Eight segments that turn responsible AI into lasting value, with accessibility, guest access and ways-of-working impact. Teams, roles, the academy and the adoption journey in this part go deeper on each.

Build capability

Leadership and sponsorship

  • Clear vision and strategic alignment
  • Senior sponsor and accountable owners
  • Cross-government collaboration
  • Adequate funding and resources
  • Visible role modelling

Build capability

Skills, learning and enablement
  • Role-based learning pathways
  • Practical, hands-on training
  • Prompts, skills and agent use
  • Protected learning time
  • Ongoing refresher content

Engage and empower

Communication and engagement
  • Clear, consistent messaging
  • What is approved, and why
  • How to get help and give feedback
  • Real examples and case studies
  • Address concerns and misinformation

Engage and empower

Champions and communities
  • Local AI champions and networks
  • Communities of practice such as xGov
  • Share prompts, skills and patterns
  • Connect pilot teams and early adopters
  • Recognise success

Support and sustain

Support and service management
  • Clear support model and service levels
  • Guidance, FAQs and how-to resources
  • Escalation for issues and incidents
  • Access to specialist expertise
  • Integration with IT and business support

Support and sustain

Measurement and value
  • Track adoption and meaningful use
  • Measure quality, productivity and outcomes
  • Include accessibility and inclusion benefits
  • Monitor cost and environmental impact
  • Report value to stakeholders

Learn and improve

Continuous improvement
  • Capture feedback and lessons
  • Share good practice across government
  • Update guidance, training and controls
  • Respond to product and model changes
  • Reassess risks and opportunities

Learn and improve

Workforce, ways of working and inclusion
  • Understand role impacts and job changes
  • Support new ways of working
  • Engage staff and representative groups
  • Promote inclusive and accessible design
  • Address capability gaps and change fatigue
Accessibility and workplace adjustments

Two questions, kept separate: is the AI tool itself accessible, and can AI support someone as a workplace adjustment? Request either through your line manager or the adjustments route. Every tool needs a non-AI alternative.

Guest and cross-tenant access

Do not rely on meeting recaps for guests. Store resources where external participants can reach them, set permissions deliberately and provide an accessible copy.

Transcription dictionary

Maintain a controlled glossary of cross-government terms and acronyms so that transcription and summaries get names right.

Employment and workplace effects

Include workload, role, delegation, autonomy, training and workplace-adjustment impact in the adoption assessment, and monitor them after rollout.

Enablers across the model

Policy and governance Approved tools and data Security and privacy Accessible by design Sustainability by design Cost control and FinOps Supplier and model management Change management Cross-tenant and external working Legal, ethical and regulatory compliance

Desired outcomes

Greater confidence
Higher productivity and quality
More inclusive services
Reduced environmental impact
Better outcomes for citizens

Ways-of-working impact questions

  • Could it replace developmental delegation or professional review?
  • Could it create meeting, message or document overload?
  • How will roles and responsibilities change?
  • What training and protected time are required?
  • How will non-users be affected?
  • What workforce evidence will be monitored?

Worked example

Worked example

Portfolio reporting grounded in the Graph.

Worked example

The same day run on Claude Projects and scheduled tasks.

Worked example

Whole-portfolio reading in one context.

Training film

The Microsoft journey, end to end.

Training film

Twin journeys film →

Both stacks, the same functions.

Comparison

Twin Capability Atlas →

39 capabilities with residency and FinOps layers.

Back to the contents · Suggest a correction (opens GitHub)

Part 8 of 12 · 9 chapters

Live pilot

Part 8 · Live pilotChapter 8.1 · 49 of 76

Live pilot handbook · prototype pilot pack, 25 August 2026

Testing this guide through a real deployment

We are not reviewing a document. We are using the playbook during a live AI deployment to determine whether it improves decisions, delivery, assurance and public-service outcomes, and whether there is sufficient evidence and appetite to establish a permanent, interactive toolkit.

Evidence status. The pilot model, thresholds and templates are proposed practice. They must be agreed for the selected live deployment. No pilot outcome is claimed in advance.

Seven stages, seven decisions

Mobilise

Is the pilot worth initiating?

Mandate, sponsor, scope and evaluation plan. Proceed or stop.

Select

Is the deployment suitable and safe for a pilot?

Candidate assessment and baseline. Select, condition or reject.

Prepare

Are people, controls, evidence and fallbacks ready?

Readiness and assurance plan. Authorise controlled deployment.

Deploy

Can the playbook be used during real delivery?

Operational evidence and decision trail. Continue, restrict, pause or stop.

Challenge

Did governance and assurance improve decisions?

Independent findings and participant evidence. Accept, remediate or retest.

Evaluate

Was the playbook useful, proportionate and valuable?

Evaluation and options appraisal. Stop, retain, simplify or evolve.

Decide

Is there appetite and evidence for a permanent capability?

Investment recommendation and roadmap. Toolkit decision.

Who this is for, and their first action

Sponsor or SRO

Decision points, risk boundaries, evidence and investment gate.

First action: Confirm outcome, authority and resourcing.
Candidate team

Eligibility, readiness criteria, expected contribution and safeguards.

First action: Complete the candidate screen.
Pilot delivery team

Step-by-step lifecycle, artefacts, ceremonies and feedback routes.

First action: Attend induction and baseline assessment.
Assurance reviewer

Review lenses, evidence standards and independence rules.

First action: Agree assurance plan and conflicts.
Interested party

Ways to observe, challenge and contribute without disrupting delivery.

First action: Choose an engagement route.
Toolkit product team

Evidence needed to decide whether to stop, retain a static guide or build a living toolkit.

First action: Establish evaluation repository.

Back to the contents · Suggest a correction (opens GitHub)

Part 8 · Live pilotChapter 8.2 · 50 of 76

Live pilot · hypothesis and test dimensions

What the pilot is testing

Primary hypothesis: using the playbook during a live deployment will improve the clarity, consistency, traceability and proportionality of governance and assurance without imposing unjustified delivery burden.

Usability

People can find and apply the right guidance without the author present.

Task observation, completion, help requests and interviews.

Proportionality

Controls reflect risk, autonomy, data, decision influence and public effect.

Exceptions, duplicated work, effort diary and reviewer judgement.

Decision quality

Material decisions have clear owners, evidence, uncertainty and conditions.

Decision records and assurance sampling.

Delivery value

The playbook prevents rework or reveals issues early enough to act.

Before and after comparison, avoided rework and issue chronology.

Assurance quality Claims are tested against meaningful evidence. Findings, closure evidence and independent opinion.

Learning value Participants gain usable capability and confidence. Applied assessments, job-aid use and follow-up interviews.

Future demand

Teams and leaders prefer a maintained toolkit over no framework or a static document.

Investment survey, sponsor decision and committed use cases.

Non-objectives

Proving that every AI deployment should use identical controls.

Treating participant satisfaction as proof of safety or value.

Using a successful pilot to claim cross-government impact.

Replacing formal legal, security, privacy, commercial or service-assessment obligations.

Back to the contents · Suggest a correction (opens GitHub)

Part 8 · Live pilotChapter 8.3 · 51 of 76

Live pilot · candidate selection

Choosing a deployment worth learning from

Seven criteria, the evidence each needs, and the condition that turns a candidate into a rejection or an escalation.

Real deployment

Named service, users, owner and intended live or limited-live use.

Reject or escalate when: There is only a hypothetical demo.

Bounded purpose

Defined tasks, prohibited uses and measurable outcomes. Reject or escalate when: Purpose is open-ended or unowned.

Safe test route

Controlled environment, data route, fallback and stop mechanism.

Reject or escalate when: Critical safeguards are unavailable.

Representative learning

The candidate exercises meaningful parts of the playbook.

Reject or escalate when: The use case is too trivial to generate evidence.

Team capacity

Product, operational, technical and governance participation.

Reject or escalate when: Delivery relies on one person or unavailable specialists.

Evidence access

Permission to examine decisions, tests, incidents and effort.

Reject or escalate when: Evaluation cannot access material evidence.

Change tolerance

Team accepts challenge and can alter or pause deployment.

Reject or escalate when: Go-live is predetermined regardless of evidence.

Recommended cohort

Use three contrasting deployments where feasible: a bounded low-risk assistant, a reusable internal knowledge or workflow agent, and a higher-impact agent influencing operational judgement. This is a suggested design, not a mandated sample.

Back to the contents · Suggest a correction (opens GitHub)

Part 8 · Live pilotChapter 8.4 · 52 of 76

Live pilot · participants and responsibilities

Seventeen roles, one accountable line

The deployment SRO stays accountable for the live service throughout. The pilot adds evaluation and independent challenge; it does not take accountability away from delivery.

Executive sponsor

Protects the purpose, secures capacity and receives the investment recommendation.

Pilot SRO Accountable for pilot outcomes, scope and residual risk.

Deployment SRO Remains accountable for the live AI-enabled service.

Pilot manager

Runs plan, ceremonies, dependencies, evidence and communications.

Product manager

Applies the playbook to needs, outcomes and backlog decisions.

Service owner

Owns operational performance, support, fallback and continuity.

User researcher

Tests usability and captures staff and affected-user evidence.

Service designer

Maps whole service, human controls, channels and failure demand.

Technical owner / AI engineer

Builds bounded capability and supplies evaluation and operational evidence.

Data and information lead

Assures sources, provenance, metadata, retention and records.

Security and privacy leads

Assess threats, access, personal data, incidents and mitigations.

Responsible AI and accessibility leads

Challenge fairness, human control, inclusion and public impact.

Learning and change lead

Delivers induction, job aids, competence assessment and adoption support.

Independent assurance lead

Plans and conducts evidence-based challenge independent of delivery.

Evaluation lead

Owns baseline, methods, synthesis, limitations and options appraisal.

Community reviewers

Provide targeted architecture, service, audit or policy challenge.

Toolkit product owner

Tests appetite and prepares the permanent-toolkit option if warranted.

Back to the contents · Suggest a correction (opens GitHub)

Part 8 · Live pilotChapter 8.5 · 53 of 76

Live pilot · formal assurance model

Five assurance stages, five kinds of opinion

Each stage states what it is for, what evidence it reviews, and the opinions it can legitimately reach.

Readiness

Determine whether controlled deployment can begin.

Evidence: Scope, owners, risk, authority, data, tests, fallback, training.

Opinion: Ready; ready with conditions; not ready.
In-flight

Test whether controls operate and the playbook supports decisions.

Evidence: Logs, samples, incidents, exceptions, user evidence, decisions.

Opinion: Effective; partially effective; ineffective.
Change

Assess material changes to model, tools, data, purpose or service.

Evidence: Impact assessment, regression tests, permissions and updated record.

Opinion: Authorise; condition; retest; reject.
Close-out

Judge whether claims and conclusions are supported.

Evidence: Outcome evidence, costs, findings, limitations and participant testimony.

Opinion: Reasonable; qualified; insufficient assurance.
Investment

Assess whether permanence is justified.

Evidence: Need, repeat demand, value, maintainability, ownership and funding.

Opinion: Stop; static guide; simplify; interactive toolkit.

Independence rule

Delivery teams own controls and evidence. Independent assurance challenges claims and reports limitations. The SRO makes the accountable decision; assurance does not transfer accountability.

Back to the contents · Suggest a correction (opens GitHub)

Part 8 · Live pilotChapter 8.6 · 54 of 76

Live pilot · journey and rhythm

End-to-end pilot journey

Ten steps from mobilise to share, each with its activities and the artefact it must produce.

Mobilise

Approve charter; appoint roles; agree scope, ethics, data and communications; define hypotheses and baselines.

Pilot charter, RACI, evaluation protocol, communication calendar.
Select

Screen candidates; interview owners; classify risk and learning value; confirm evidence access.

Candidate scorecard, selection decision and conditions.
Induct

Brief all roles; assess learning needs; issue playbook routes and job aids; rehearse escalation.

Induction record and competence baseline.
Baseline

Capture current governance journey, effort, uncertainty, rework, readiness and expected outcomes.

Baseline map and measure definitions.
Apply

Use playbook lifecycle, authority model, internal ATRS record, evidence pack and decision gates.

Completed artefacts and traceable decisions.
Deploy

Begin bounded live or limited-live operation; monitor; sample outputs; record exceptions and effort.

Operational evidence, incidents, feedback and change log.
Challenge

Run architecture, service, security, privacy, responsible-AI, accessibility and audit reviews.

Formal findings and closure plan.
Evaluate

Compare baseline; triangulate metrics, observation and interviews; document limitations.

Evaluation report and playbook backlog.
Decide

Options appraisal and sponsor decision: stop, static, simplify or toolkit.

Decision record, investment recommendation and roadmap.
Share

Return findings to participants and communities; publish only what is approved and evidenced.

Lessons pack and controlled reusable assets.

Ceremonies and evidence rhythm

Candidate interview

Pilot, deployment and assurance leads.

Test eligibility, constraints and evidence access. Selection record.

Kick-off and induction

All core roles.

Align purpose, boundaries, responsibilities and stop routes.

Signed charter and readiness actions.
Weekly evidence clinic

Pilot manager, product, operations, evaluation.

Review decisions, friction, incidents, metrics and learning.

Evidence log and backlog.
Fortnightly assurance checkpoint

SRO, service owner, assurance specialists.

Test controls and conditions; decide remedial action. Assurance note and decision.

Participant pulse

Users and delivery team.

Capture usability and burden close to the work. Short structured feedback.

Mid-point challenge

Independent reviewers and community representatives.

Identify blind spots before conclusions harden. Challenge findings.

Close-out review

All accountable and assurance roles.

Agree evidence, limitations and recommendations. Final assurance opinion.

Investment gate

Sponsor, SRO, toolkit owner, finance and governance.

Choose future option. Investment decision.

Back to the contents · Suggest a correction (opens GitHub)

Part 8 · Live pilotChapter 8.7 · 55 of 76

Live pilot · communications plan

Nine moments, one honest message

Each moment names its audience, message, channel and owner. The framework is being assessed — not the teams taking part.

Pre-launch

Sponsors and governance

Why this is a live deployment test; decision sought; safeguards.

Executive brief and sponsor conversation. Pilot sponsor.

Recruitment
Candidate teams

The framework is being assessed, not the team; eligibility and contribution.

Teams post, briefing session and candidate pack. Pilot manager.

Selection
Applicants

Transparent criteria, result, conditions and alternative engagement.

Decision note and direct conversation. Pilot SRO.

Induction
Selected participants

Process, roles, boundaries, evidence, learning and escalation.

Workshop, handbook and role cards. Learning lead.

In-flight
Participants and sponsors

Progress, decisions, risks, changes and support.

Weekly digest and monthly dashboard. Pilot manager.

Challenge
Assurance and community

Focused review question, evidence boundaries and response route.

Review pack and facilitated session. Assurance lead.

Close-out
All participants

What evidence shows, limitations and what will change.

Playback and findings report. Evaluation lead.

Investment
Senior decision-makers

Options, value, demand, costs, ownership and recommendation.

Decision paper and deck. Pilot SRO.

Wider sharing
Interested communities

Verified lessons, reusable patterns and next decision.

Community session and controlled publication. Communications owner.

Core messages

This is a controlled test of the framework in real delivery.

Participation is not an audit of team performance.

Safety, service obligations and accountable decisions override pilot learning.

Evidence includes what fails, adds burden or duplicates existing controls.

No decision to establish a permanent toolkit has been made in advance.

Back to the contents · Suggest a correction (opens GitHub)

Part 8 · Live pilotChapter 8.8 · 56 of 76

Live pilot · success framework and feedback

Twelve measures, no invented thresholds

Each measure names its evidence, owner and cadence. Thresholds are agreed after the baseline is captured: a pre-set percentage without context would create false precision.

Independent use

Participants navigate lifecycle and required artefacts without author dependency.

Observed task completion; help requests; navigation errors. Evaluation lead. Baseline and end.

Proportionality

Teams and reviewers agree controls match material risk and consequence.

Effort diary; duplicated controls; exceptions; reviewer judgement.

Assurance lead.

Weekly and close-out.

Decision traceability

Material decisions identify owner, evidence, uncertainty, conditions and review.

Decision-record sample. Pilot manager. Fortnightly.

Early issue detection

Material gaps are discovered while action remains possible.

Issue chronology and avoided or reduced rework evidence. Service owner. Monthly.

Assurance quality

Findings distinguish intent, design, implementation, operation and outcome.

Independent assurance quality review. Assurance lead. Mid-point and close.

Operational readiness

Support, monitoring, fallback and incident routes work in exercises and operation.

Readiness checks, exercises and operational records. Operations lead. Pre-live and live.

Learning efficacy

Participants demonstrate relevant competence in realistic tasks.

Applied assessment and coaching records. Learning lead. Before and after.

User and service value

Observed service outcomes improve or credible learning justifies continuation.

Service-specific baseline and evaluation. Product manager. Agreed cadence.

Burden and cost

Implementation effort and recurring cost are visible and acceptable relative to value.

Time, support and consumption data. Evaluation and finance. Monthly.

Demand and appetite

Leaders and teams prefer a maintained option and commit use cases, ownership or resources.

Structured option choice and commitments. Toolkit product owner. Close-out.

Evidence integrity

Claims separate verified results, internal evidence, assumptions and expected outcomes.

Evidence audit. Independent assurance. Close-out.

Inclusion

Research and controls address accessibility and affected-user differences.

Accessibility and inclusion evidence. Accessibility lead. Design and close-out.

Feedback as formal evidence

Observed use Whether participants can apply the guide in context. At key lifecycle tasks. Observe the framework, not individual performance.

Decision sampling Whether governance claims are traceable. Throughout deployment. Risk-based sample agreed in advance.

Effort diary Where burden, duplication and rework occur. Weekly. Record estimates honestly and label them.

Structured interviews What helped, hindered or remained missing. Baseline, mid-point, close. Separate experience from factual claims.

Tabletop scenarios Whether controls work under failure and uncertainty. Pre-live and mid-point. Use synthetic scenarios where appropriate.

Assassin review

Where the model could fail, be gamed or become bureaucratic.

Mid-point.

Facilitated, blameless challenge.

Independent evidence review Whether conclusions follow from evidence. Close-out. Declare conflicts and evidence limitations.

Option-preference survey Whether appetite exists for future forms. Close-out. Require rationale and commitment, not preference alone.

Community consultation Whether patterns transfer beyond the pilot. After initial evidence. Do not generalise from advocacy alone.

Feedback taxonomy

KEEP

Works and should remain.

Protect and evidence.

IMPROVE

Useful but unclear or incomplete.

Refine and retest.

REMOVE

Adds unjustified burden or duplicates controls.

Remove or explain exception.

MISSING

A necessary need or control is absent.

Add, assign owner, validate.

CONFLICT

Contradicts policy, service or assurance practice.

Escalate and resolve.

LOCALISE

Valid but needs context-specific adaptation.

Create configurable pattern.

Back to the contents · Suggest a correction (opens GitHub)

Part 8 · Live pilotChapter 8.9 · 57 of 76

Live pilot · decision and future state

What happens after the pilot

Four options, the conditions that select each, and what each commits the organisation to. No option is pre-selected.

Stop

Value is weak, burden is excessive, risks are unresolved or demand is absent.

Archive evidence and retain lessons; do not operationalise.

Retain static playbook

The guide is useful as reference but repeat operational demand is limited.

Assign light-touch ownership and scheduled review.

Simplify and retest
  • Value exists but usability or proportionality is inadequate.
  • Prioritise critical changes and conduct focused retest.
Build living toolkit

Evidence shows repeat need, practical value, appetite and sustainable ownership.

Fund discovery and product development for an ongoing service.

Investment evidence required for option D

Multiple committed user groups or deployments, not general interest alone.

A named product or service owner and multidisciplinary maintenance capacity.

A sustainable source of policy, standards and pattern updates.

Clear content governance, versioning, supersession and retirement rules.

Evidence that dynamic interaction adds value beyond a downloadable guide.

Security, privacy, accessibility, information-management and operating-cost assessment.

A measurable service outcome and roadmap, not only a feature list.

Future-state toolkit concept

Flagship implementation guide

Understand the end-to-end route.

Adaptive journey through lifecycle, roles, gates and evidence.

Evidence: Navigation, completion and repeat-use data.

Governance navigator

Know which controls and forums apply.

Context-led route based on purpose, risk, autonomy and public effect.

Evidence: Routing errors, exceptions and decision consistency.

Internal ATRS workspace

Create and maintain a transparency record.

Guided record, evidence links, review and publication decision.

Evidence: Field usability, maintenance burden and scope decisions.

Assurance workspace

Build and review an evidence-based assurance case. Claims, evidence, findings, conditions and closure. Evidence: Evidence quality and review efficiency.

Learning academy

Build competence by responsibility. Role pathways, scenarios, assessment and job aids. Evidence: Competence outcomes and support demand.

Pattern library

Reuse proven controls and artefacts. Versioned patterns with evidence, owner and applicability. Evidence: Verified reuse and reduced duplication.

Portfolio control tower

Oversee value, risk, cost and lifecycle. Inventory, status, evidence freshness and decision queues. Evidence: Decision usefulness and data quality.

Community feedback loop

Keep the toolkit relevant.

Structured proposals, challenge, supersession and release notes.

Evidence: Contribution quality and maintenance capacity.

Product principle: the implementation guide remains the flagship resource. Interactive functions should reduce cognitive load and help people act safely; they must not hide accountability, evidence or applicable authoritative guidance.

Participant checklist

Before deployment
  • Read assigned playbook route.
  • Complete induction and competence tasks.
  • Confirm purpose, authority and prohibited use.
  • Validate data, tools and owners.
  • Practise fallback and escalation.
During deployment
  • Use approved process and boundaries.
  • Record material decisions and uncertainty.
  • Report incidents and friction promptly.
  • Maintain evidence and effort diary.
  • Attend evidence clinics and assurance reviews.
At close-out
  • Complete interview and option survey.
  • Review evidence and factual accuracy.
  • Declare unresolved concerns.
  • Identify KEEP, IMPROVE, REMOVE or MISSING.
  • Confirm future interest and commitment.

Pilot safety stop rules

  • Immediate or plausible material harm to users or the public.
  • Loss of control over permissions, data or state-changing actions.

Critical security, privacy, legal or accessibility condition breached.

Required human oversight or operational fallback unavailable.

Evidence integrity compromised in a way that prevents accountable decision-making.

Deployment moves outside approved purpose or environment.

Quick templates

Candidate screen
  • Service and user need
  • Agent purpose and prohibited use
  • Users and affected parties
  • Data, tools and authority
  • Expected public-service influence
  • Named SRO, service and technical owners
  • Safe test environment and fallback
  • Evidence access and team capacity
  • Why this candidate creates transferable learning
Weekly evidence log
  • Date
  • Decision or event
  • Evidence
  • Effect on delivery
  • Playbook feedback
  • Owner and action
Close-out option record
  • Option
  • Evidence for
  • Evidence against
  • Conditions
  • Decision owner

Source and status map

AI Playbook for the UK Government

Human control, lifecycle management, governance, skills, quality assurance and risk.

Authoritative external grounding.
Service Standard and Service Manual

User needs, multidisciplinary delivery, security, measures, reliability and improvement.

Applied proportionately as a proposed service lens.
Technology Code of Practice

Criteria for government technology, including user needs, accessibility, security, privacy, data and Service Standard alignment.

Authoritative external grounding.

ATRS guidance and scope policy Transparency purpose, scope tests and publication route. Official process; internal registry extension is proposed.

Responsible AI Toolkit and AI Knowledge Hub

Responsible deployment, assurance, governance and practical resources.

Authoritative and current resource grounding.
AI Agents as Digital Workers Playbook

Policy, lifecycle, roles, internal registry, learning and roadmap prototype.

Primary prototype under live test.

Final discipline. Pilot conclusions must distinguish observed and verified results, internally evidenced results, participant views, assumptions, expected outcomes and unresolved unknowns.

Back to the contents · Suggest a correction (opens GitHub)

Part 9 of 12 · 8 chapters

Product documentation

Part 9 · Product documentationChapter 9.1 · 58 of 76

Delivery documentation · PRD

Product requirements

What this guide is for, who it serves, what it must do and what it deliberately does not do. Derived from published guidance and community discussion; a proposal, not an adopted product mandate.

Problems worth solving

Each problem is visible in the source material rather than assumed.

Guidance is real but scattered

The AI Playbook, Service Standard, TCoP, ATRS, NCSC secure-AI guidelines, ethics frameworks and FinOps practice exist separately. Teams reassemble them from scratch, inconsistently, each time.

Experimentation outruns governance

Personal assistants become shared agents and then business-critical dependencies without registration, named ownership, assurance or funded support.

Cost is discussed as licences

Token and licence spend is visible; data remediation, workflow redesign, assurance, human review, monitoring, support and exit are not costed or owned.

Evidence is claimed, not demonstrated

The same product yields different results under different evaluation methods, so headline productivity figures travel further than the method that produced them.

Community knowledge decays

Sessions, transcripts, repositories and personal networks hold the learning; there is no shared discovery layer, taxonomy or evidence pack, so lessons are relearned.

Goals and how each is judged

One route from mandate to governed operation

A reader can go from no policy position to a registered, assured, monitored agent using only this guide.

All eight start-here stages have a chapter and a tool
Make the invisible costs visible

Every cost class in the investment line is named, owned and modelled.

Cost lab produces a per-run and monthly figure with review effort included
Proportionate assurance, not blanket control

Criticality drives the gate; low-risk work stays fast.

Five tiers with stated evidence and review frequency
Evidence before assertion

Claims carry method, publisher and date; illustrative material is labelled.

Every figure in the guide is attributed or marked illustrative
Reusable artefacts, not prose

Register schema, ATRS record, instruction block and evidence card can be copied out and used.

Four artefacts exportable or copyable
Accessible to every role

Leaders, engineers, assurance, commercial, finance and frontline staff each have a path.

Role pathways cover the twelve priority roles
In scope

Guidance synthesis and chapter navigation

Assessment tools (readiness, criticality, tiers, maturity, quality gate)

  • Register and ATRS record schemas
  • Cost and unit-economics modelling
  • Skill instruction builder
  • Evidence, use-case and international comparison libraries
  • Illustrative portfolio dashboard
  • Delivery documentation set (this part)
Out of scope
  • Holding the authoritative record of any live deployment
  • Storing personal or OFFICIAL-SENSITIVE data
  • Connecting to tenant telemetry, licensing or billing systems
  • Issuing approvals or acting as the assurance authority
  • Publishing ATRS records on behalf of an organisation
  • Replacing or overriding central or departmental guidance

Non-functional requirements

Accessibility WCAG 2.2 AA: contrast, focus visibility, keyboard operation, reduced-motion, semantic headings and labels.Information handling OFFICIAL working content only. No personal data, no case data, no credentials. Assessment answers stay on the reader's device.Portability Runs as a single file in a browser with no build step, no backend and no network dependency beyond fonts and cited links.
Performance First paint before the whole document streams; interactions respond within one frame budget on a standard civil-service laptop.Provenance Crown-copyright content reused under the Open Government Licence v3.0, with publisher and date on every figure.Print and export Chapters print legibly; the register schema and instruction block copy to the clipboard as plain text.

Release phases

Now

Phase 1 — proposal

This guide, reviewed by the community, with illustrative figures clearly labelled.

Next

Phase 2 — validated

Illustrative figures replaced with measured departmental data; international benchmarking executed.

Later

Phase 3 — adopted

Referenced in departmental AI policy, with the register schema and assurance gates in operational use.

Risks to the product itself

Illustrative figures read as measured High

Label every illustrative frame; keep publisher and date on evidence; publication checkpoint before circulation.

Guide treated as authority Medium

Proposal status on the landing chapter and in the hierarchy; explicit statement that it does not override guidance.

Content drifts from source guidance High

Chapter-to-source map; review date per chapter; ticker tracks guidance in force.

Assessment answers mistaken for assurance Medium

Device-only storage, no scoring authority claimed, evidence-based maturity rather than self-scored levels.

Back to the contents · Suggest a correction (opens GitHub)

Part 9 · Product documentationChapter 9.2 · 59 of 76

Delivery documentation · FTIR specification

Functional, technical and integration requirements

Numbered requirements with priority, traceable to the chapters that satisfy them. Must, should and could follow the usual prioritisation convention.

Functional

FR-01

Navigate 69 chapters in eleven numbered parts with contents, previous, next and sibling links

Must

FR-02

Search every chapter, principle, gate, register field, role and source from one field

Must

FR-03 Record which chapters have been opened and show progress Should

FR-04

Rate readiness across eight dimensions and return a banded barometer

Must

FR-05 Determine an assurance tier from a two-question intake Must

FR-06

Score agent criticality across ten dimensions and derive the gate

Must

FR-07 Score barriers by impact and likelihood and band the result Should

FR-08

Model cost per run and per month across models, languages and volumes

Must

FR-09

Model unit economics: value per interaction, net value and return

Must

FR-10 Build a seven-part instruction block and copy it out Must

FR-11

Assess evidence strength per capability domain across seven maturity levels

Should

FR-12 Step a failure and recovery loop, marking stages complete Should

FR-13

Present the eighteen-field register schema and the eight-section ATRS record

Must

FR-14

Show an illustrative portfolio dashboard with lifecycle, assurance, risk and domain views

Could

Technical

TR-01

Single-file browser document; no build step, bundler or server runtime

Must

TR-02

All state in memory; only the readiness barometer persists, in localStorage on the reader's device

Must

TR-03 Inline styling so content paints as it streams Must

TR-04

Keyboard access for every control; visible focus; reduced-motion honoured

Must

TR-05

Contrast at or above 4.5:1 for text and 3:1 for meaningful non-text

Must

TR-06 No third-party analytics, cookies or tracking Must

TR-07

External links open in a new tab with noopener and carry publisher and date

Should

TR-08

Deterministic calculations — same inputs, same outputs, no model calls

Must

Integration

IR-01

Register schema aligns field-for-field with the ATRS record sections

Must

IR-02

Register, information asset register and data catalogue treated as three integrated registers, not duplicates

Must

IR-03

Criticality score maps to the five assurance gates and their evidence requirements

Must

IR-04

Cost model outputs feed the investment line and FinOps allocation layers

Should

IR-05 Instruction block output is platform-neutral plain text Must

IR-06 Chapters cite the guidance in force and carry a review date Must

IR-07

Links out to the AI Engineering Lab, AI Knowledge Hub and departmental policy rather than copying them

Should

IR-08

Sits alongside the govukOS twin front door and portals as one app in a shared shell

Should

Back to the contents · Suggest a correction (opens GitHub)

Part 9 · Product documentationChapter 9.3 · 60 of 76

Delivery documentation · SAD

Solution architecture

Principles, context, layers and the decisions that shaped them. The architecture is deliberately small: the guide holds interpretation, never the record of a live deployment.

Architectural principles

Guidance is referenced, not copied

Link to the governed source; hold interpretation and sequence here.

No shadow record

The guide never becomes the register of record for a live deployment.

Client-side by default

No backend means no new data controller, no new attack surface, no new service to run.

Accessible before decorative

Contrast, focus and keyboard behaviour are acceptance criteria, not polish.

Evidence carries method

Every figure states publisher, date and method, or is labelled illustrative.

Open by default

Crown-copyright content under OGL v3.0; the guide itself is reusable and adaptable.

Context — actors and their part

Reader Civil servant in one of the twelve priority roles Reads chapters, runs assessments, copies artefacts

Owning organisation Department, agency, council or health body Holds the register of record, policy and approvals

Central guidance GDS, DSIT, NCSC, CDDO successor functions Publishes the guidance this guide interprets

Community

Cross-government Copilot forum, AI Engineering Lab, ethics and data communities

Supplies patterns, failures and supplier intelligence

Suppliers Platform, model and integration providers Subject of dependency, exit and commercial chapters

govukOS shell Twin front door and sibling portals Hosts the guide as one app among several

Logical layers

Experience

Masthead, ticker, drawer, chapter rail, contents, chapter body, footer navigation

Chapter registry Parts, chapters, numbering, accents, icons, read state

Content

Source-derived data: principles, gates, domains, roles, tiers, schemas, dashboards

Assessment logic

Barometer, criticality, tiers, maturity evidence, quality gate, barrier scoring

Economics

Token maths, language multipliers, unit economics, allocation layers

Local state

In-memory interaction state plus one persisted barometer record on the device

Decisions

ADR-01

Single-file client-side document

No hosting, no data processing agreement, no service to operate; trades server-side features for portability.

ADR-02

No personal or case data anywhere

Removes the need for a DPIA on the guide itself; assessments stay on the device.

ADR-03

One artefact, many chapters

Keeps navigation, search and cross-references coherent; avoids a fragmented document set.

ADR-04

Illustrative data labelled, never silently synthetic

Protects the evidence boundary this playbook sets out.

ADR-05

Accent colour per part

Gives readers a positional cue; all accents are contrast-tested as text.

Back to the contents · Suggest a correction (opens GitHub)

Part 9 · Product documentationChapter 9.4 · 61 of 76

Delivery documentation · HLD

High-level design

Components, the entities they work on, and the four flows that carry most of the value.

Components

Shell

Masthead, guidance ticker, launchpad, search field

Holds identity and cross-app links
Navigation

Drawer with parts and chapters, chapter rail, contents overlay, prev/next

Owns chapter numbering and read state
Chapter renderer

One section per chapter, streamed in order

Eyebrow, accent title, lede, body blocks
Assessment engine

Barometer, criticality radar, tier finder, maturity evidence, quality gate

Pure functions over local state
Economics engine

Token cost, language multiplier, unit economics, allocation layers

Deterministic arithmetic, GBP formatting
Artefact builders

Instruction block, register schema, ATRS record, evidence card

Copy-to-clipboard output
Search index

Flat index of chapters, principles, gates, fields, roles, sources

Substring match with type and destination
Dashboard

Illustrative portfolio control tower

Read-only, clearly labelled illustrative

Entities

Chapter id, part, number, title, icon, accent, read Navigation and progress

Blueprint title, source, ingest date, chapters it feeds Provenance and coverage

Assessment dimension, score, band, timestamp Barometer and maturity evidence

Register record 18 fields from identity to retirement The record an organisation holds

ATRS record 8 sections, each with its own fields Transparency record per agent

Cost scenario tokens in/out, prices, tools, infra, review effort, volume Cost and unit economics

Instruction block goal, context, sources, task, format, guardrails, output Skill injection

Key flows

Read a chapter Drawer or contents → chapter renders → read state recorded → next chapter offered

Assess readiness Rate eight dimensions → band computed → stored on device → shown in the drawer workspace

Tier an agent Two intake questions → tier → gate, controls, evidence and review frequency shown

Price a job Enter words in and out, model, language, volume → per-run and monthly cost → unit economics

Back to the contents · Suggest a correction (opens GitHub)

Part 9 · Product documentationChapter 9.5 · 62 of 76

Delivery documentation · LLD

Low-level design

Module responsibilities, the arithmetic behind every score, the validation rules and the accessibility implementation. Enough for someone else to rebuild the behaviour exactly.

Modules and their state

Navigation module

groupDefs → flat chapter list → numbering, accents, prev/next, contents

section, groups, read, toc

Barometer module eight dimension scores → mean → band with colour and label scores (persisted)

Criticality module ten dimension scores → total → tier and gate mapping crit, gateTier

Barrier module impact × likelihood → 1–25 → four bands bar

Economics module

tokens = words ÷ 0.75; cost = in×priceIn + out×priceOut + tools + infra + review

lab, ue

Instruction module seven fields → composed block → clipboard blk, copied

Algorithms

Barometer banding

mean of rated dimensions; 0–0.9 Nascent, 1.0–1.9 Emerging, 2.0–2.4 Establishing, 2.5–3.0 Leading; unrated dimensions excluded from the mean and reported as unrated.

Criticality to gate

sum of ten dimensions scored 1–5; 10–20 personal, 21–30 team, 31–38 departmental, 39–44 operational service, 45–50 citizen impact.

Barrier banding

impact × likelihood on 1–5 scales; 1–3 low, 4–8 medium, 9–15 high, 16–25 critical.

Unit economics

value per interaction = (minutes saved ÷ 60 × hourly rate) × quality factor; net = value − cost; return = net ÷ cost expressed as a percentage.

Validation rules

Numeric inputs Clamp to declared minimum and maximum; reject non-numeric silently rather than throwing.

Unrated dimensions Show an em dash, never a zero, and exclude from means.

Currency Two decimal places, pound sign, thousands separated; never rounded away in totals.

Clipboard Plain text only; confirm with a transient state, no alert.

Persistence One key, versioned; parse defensively and fall back to blank on any error.

Accessibility implementation

Colour Every text and meaningful non-text pair contrast-tested; accents have separate text-safe variants where a fill carries text.

Keyboard All controls are buttons or inputs; aria-pressed and aria-current reflect state; focus ring never suppressed.

Motion The guidance ticker pauses on hover and respects prefers-reduced-motion.

Structure One h1 per chapter, sequential headings, aria-labels on sections, skip link to content.

Language British English, plain terms, expanded acronyms on first use in a chapter.

Back to the contents · Suggest a correction (opens GitHub)

Part 9 · Product documentationChapter 9.6 · 63 of 76

Delivery documentation · theory of change

Theory of change

From inputs to impact, with the assumptions each step rests on. Written so the chain can be falsified: if an assumption fails, the link it supports fails with it.

Inputs

Central guidance in force Community evidence and patterns Evaluation studies Departmental policy and delegated authority Funded time from named roles

Activities

Synthesise guidance into sequenced chapters Build assessment and costing tools Publish register and ATRS schemas Run role-based capability pathways Return findings to the community

Outputs

One navigable implementation guide Completed readiness and criticality assessments Registered agents with named owners Costed business cases including hidden costs Reusable evidence cards and patterns

Short-term outcomes

Teams find the applicable guidance quickly Proportionate assurance replaces blanket caution Agents have owners, tiers and review dates Cost conversations include people and exit

Medium-term outcomes

Fewer orphaned and unregistered agents Duplicated procurement and assurance effort falls Evidence is reused across organisations Oversight is continuous, not one-off

Impact

Governed AI-enabled services that create demonstrable public value
Sustained public trust in how government uses AI
Each row depends on the one above it holding. Outcomes are claimed only where evidence exists; everything below outputs is currently a proposal.

Assumptions

Guidance stays reasonably stable Chapters carry review dates so drift is visible rather than silent.

Teams have protected time Without funded time, the guide is read but not applied.

Leaders accept proportionality Blanket caution defeats the tiering model as surely as no control at all.

Community keeps sharing Reuse depends on organisations publishing failures as well as successes.

Illustrative figures get replaced Measured departmental data must supersede the labelled placeholders.

Indicators to track

Chapters opened per role pathway

Agents registered with a named owner and review date

Proportion of agents with an ATRS record

Assurance decisions made at the tier they belong to

Business cases including the full investment line

Patterns and evidence cards reused across organisations

Back to the contents · Suggest a correction (opens GitHub)

Part 9 · Product documentationChapter 9.7 · 64 of 76

Delivery documentation · user research bundle

Personas, stories, journeys and service blueprints

Drawn from the roles, frictions and lifecycle material in published guidance and community discussion. These are evidence-derived working personas, not research participants — they need validating with real users before they carry design decisions.

Six personas

Accountable leader
Director or SRO, accounting-officer line

Decide where AI is worth the risk and the money, and be able to answer for it.

Needs

  • Evidence that a claim is measured, not asserted
  • A defensible risk appetite
  • One page that survives a select-committee question

Frustrations

  • Productivity figures with no method attached
  • Being asked to approve what nobody owns
  • Cost framed as licences only
Reads: Assurance tiers, evidence chapter, investment line, control tower
Service owner
Owns an operational service with real users

Adopt agents without destabilising a service people depend on.

Needs

  • Clear promotion gates
  • Named ownership and support model
  • A rollback and exit route

Frustrations

  • Agents appearing in the service without registration
  • Unclear escalation when output is wrong
  • Support cost landing on the service unfunded
Reads: Agent journey, criticality and gates, failure and recovery, dependency and exit
AI engineer
Builds and evaluates agents

Build well-grounded agents and prove they work.

Needs

  • Evaluation and red-team patterns
  • Grounding decision tree
  • Instruction blocks that transfer between platforms

Frustrations

  • Ungoverned copies of source data made for grounding
  • No baseline to evaluate drift against
  • Prototype standards applied to production, and the reverse
Reads: Skill studio, architecture and quality, knowledge foundation, AI Engineering Lab
Assurance and ethics specialist
Reviews, challenges and signs off

Apply proportionate control and see the evidence behind it.

Needs

  • Criticality scoring that distinguishes usage from consequence
  • ATRS record completeness
  • A residual-risk statement per agent

Frustrations

  • Assurance as a gate at the end rather than a thread throughout
  • Self-scored maturity with no evidence
  • Incidents surfacing without a route to redress
Reads: Risks and mitigations, internal ATRS registry, maturity evidence, autonomy boundaries
Commercial and FinOps lead
Buys, allocates and forecasts

Fund adoption honestly and keep exit possible.

Needs

  • Unit economics per use case
  • Allocation model across six cost layers
  • Supplier requirements for telemetry, portability and exit

Frustrations

  • Token cost mistaken for total cost
  • Lock-in discovered at renewal
  • Shared costs with no agreed allocation
Reads: Cost value and investment, commercial and suppliers, FinOps, dependency and exit
Frontline practitioner
Uses agents in daily casework

Do the work faster without carrying unowned risk.

Needs

  • Plain guidance on what is allowed
  • A way to challenge an output
  • Training in the flow of work

Frustrations

  • Guidance written for architects
  • No clear escalation when the agent is wrong
  • Being blamed for a tool decision made elsewhere
Reads: Support around every colleague, inclusive adoption, roles and pathways

User stories with acceptance criteria

As a Accountable leader, I need to see the method behind any productivity claim, so that I do not repeat an unevidenced figure in public.

Accepted when: Every figure shows publisher, date and method, or is labelled illustrative

As a Accountable leader, I need to know which agents are high risk and who owns them, so that I can direct attention where consequence is highest.

Accepted when: Control tower lists risk level, owner and next review per agent

As a Service owner, I need to know the gate an agent must pass before it touches my service, so that I can plan assurance rather than discover it.

Accepted when: Criticality score returns a tier with named evidence and review frequency

As a Service owner, I need to have an exit path for a platform or model, so that a supplier change does not stop my service.

Accepted when: Seven-step exit path with the dependencies it addresses

As a AI engineer, I need to decide where grounding information should live, so that I do not create ungoverned copies.

Accepted when: Decision tree returns one of four options in priority order

As a AI engineer, I need to compose an instruction block once and reuse it, so that behaviour is consistent across platforms.

Accepted when: Seven-part block copies out as plain text

As a Assurance specialist, I need to score criticality across ten dimensions, so that control is proportionate to consequence.

Accepted when: Score maps to one of five gates with its controls

As a Assurance specialist, I need to record transparency information for every internal agent, so that oversight and audit have something to inspect.

Accepted when: Eight-section ATRS record with all fields listed

As a FinOps lead, I need to model cost per run and per month, so that a business case includes consumption, not guesses.

Accepted when: Cost lab returns per-run and monthly figures with a language multiplier

As a FinOps lead, I need to see value against cost per interaction, so that I can show whether the agent earns its keep.

Accepted when: Unit economics returns value, net and return

As a Frontline practitioner, I need to know what I may use and when to escalate, so that I can work confidently within the guardrails.

Accepted when: Support chapter states allowed use, escalation and where help comes from

As a Community manager, I need to turn a session into a reusable evidence pack, so that learning survives the meeting.

Accepted when: Evidence pack structure with owner, taxonomy and review date

Journeys

First-time reader to first decision

Arrives with a question Lands on the overview, sees proposal status “Is this authoritative?” Proposal flag and hierarchy answer it in one screen

Finds the applicable chapter Uses search or contents “Where does my case sit?” Chapter numbering and part accents orient quickly

Runs an assessment Tier finder or criticality radar “Am I over- or under-controlling?” Tier arrives with evidence and review expectations

Takes an artefact away Copies the register schema or instruction block “What do I do on Monday?” Copyable artefacts, not prose

Returns for oversight Re-opens the control tower and maturity evidence “Has anything drifted?” Read state and review cadence bring them back

Agent from idea to retirement

Idea Someone builds a helpful agent “This saves me an hour a day” Register at creation, not at scale

Share Colleagues start relying on it “It works, so why formalise it?” Promotion gate between personal and team scope

Assure Criticality scored, tier applied “How much control is proportionate?” Gate names the evidence needed

Operate Monitoring, cost and quality watched “Who notices if it degrades?” Named owner and continuous duties

Review Re-authorise, improve, restrict or retire “Is it still earning its authority?” Review date forces the decision

Retire Access revoked, records archived “What happens to the evidence?” Archive and audit stage in the lifecycle

Community session to reused pattern

Session happens Weekly call, recorded from 17 August 2026 “Will I find this again?” Automatic recording closes part of the capture gap

Capture Transcript, chat, files, commitments “Who owns the follow-up?” Session evidence pack with owner and actions

Classify Taxonomy: capability, use case, sensitivity, owner, review date “Which of these applies to me?” Shared taxonomy across communities

Connect Federated index links to governed sources “Is this the current version?” Index links rather than copies

Reuse Another organisation adopts the pattern “Has anyone tried this already?” Discovery by role, sector, maturity and supplier

Learn Outcome and corrections returned “Did it actually work?” Value reported as reuse, not attendance

Service blueprint — an agent in an operational service

Five lanes across the same five moments: what the user does, what they see, what happens behind it, what supports it, and what is recorded.

User actions

Requests help Reviews agent output Accepts or challenges Escalates an error Feeds back on value

Frontstage

Agent responds with citation Confidence and limits shown Human review point Escalation route offered Feedback captured in service

Backstage

Retrieval over authoritative sources Guardrails and permission checks Logging and audit trail Incident triage and rollback Evaluation against baseline

Supporting processes

Register and ATRS record maintained Assurance gate evidence held Cost allocated to the use case Supplier and model change managed Review or retirement decision

Evidence and records

Source provenance Approval and sign-off Monitoring and drift reports Incident records Benefit evidence

Back to the contents · Suggest a correction (opens GitHub)

Part 9 · Product documentationChapter 9.8 · 65 of 76

Delivery documentation · external resources

The AI Engineering Lab and other gov.uk resources

This guide interprets and sequences; these resources supply engineering practice, worked examples and published evaluation. Link to them rather than copying them — they move faster than any document.

AI Engineering Lab

GDS Office of the CTO engineering-excellence project with DSIT

Programme site, an open GitHub resource repository under OGL v3.0, a Knowledge Hub community, role-specific material and cross-government hackathons. Enquiries to the GDS engineering-excellence team.

Use it for: Engineering patterns, evaluation practice, contribution route
Open resource
AI Knowledge Hub
i.AI, GDS within DSIT

A living resource on building and adopting AI in the public sector, with tool pages carrying what was built, what it cost and what the evaluation found.

Use it for: Worked examples with evidence, before you build your own
Open resource
i.AI product family
Incubator for Artificial Intelligence

Caddy, Consult, Extract, Redbox, Minute, Parlex, Scout, Connect, Lex and rAPId — several open-sourced, with published evaluations.

Use it for: Reusable patterns and honest post-mortems, including a sunsetted product
Open resource
AI Growth Lab
DSIT

Proposed cross-economy regulatory sandbox with time-limited exemptions under supervision; call for evidence closed January 2026.

Use it for: Where regulation, not capability, is the blocker
Open resource
Fundamental AI Research Lab
DSIT with UKRI

Up to £40m for long-term research into limitations such as hallucination, unreliable memory and unpredictable reasoning.

Use it for: Why some current risks are structural, not configuration errors
AI Playbook for the UK Government
GDS, February 2025

Ten principles and the life cycle this guide sequences; supersedes the Generative AI Framework.

Use it for: The guidance in force
Open resource

Which resource, which chapter

Need an engineering or evaluation pattern AI Engineering Lab repository and community 3.2 Architecture & quality

Need a worked example with published evidence AI Knowledge Hub tool pages 2.2 Use cases & evidence

Need to see how a product was evaluated i.AI evaluations and blogs 2.1 Evidence before assertion

Need to know if regulation is the blocker AI Growth Lab call for evidence Resource library (home page)

Need the authoritative principles AI Playbook for the UK Government 1.3 The playbook

Need transparency-record structure ATRS and the internal registry chapter 4.5 Internal ATRS registry

Back to the contents · Suggest a correction (opens GitHub)

Part 10 of 12 · 4 chapters

Knowledge and navigation

Part 10 · Knowledge and navigationChapter 10.1 · 66 of 76

Knowledge and navigation · 10 entries

Thesaurus

Preferred term, the variants you will hear, and why the distinction matters. Cross-government acronyms and loose synonyms are the main cause of transcripts and registers that cannot be searched.

Agent

digital worker

assistant

copilot

bot

Use agent for the governed capability; digital worker only when the employment framing is deliberate.

Assurance

governance

compliance

sign-off

Assurance tests claims against evidence; governance sets the rules. Do not use them interchangeably.

Criticality

risk level

importance

priority

Criticality is dependency and consequence. Risk level is the output of scoring it.

Register

inventory

catalogue

asset register

Register for agents; catalogue for data; asset register for information assets. Three registers, integrated, not merged.

Evaluation

testing

benchmarking

evals

Evaluation covers accuracy, safety and fitness against a baseline; testing alone implies pass or fail.

Guardrail

control

policy

limit

Guardrails are technical constraints; controls include process and people; policy is the rule they enforce.

Retirement

decommissioning

offboarding

sunsetting

Retirement is the lifecycle stage; decommissioning is the technical work inside it.

Transparency record

ATRS record

algorithmic transparency

publication

ATRS record is the published artefact; the internal registry extension is proposed, not mandated.

Value

benefit

ROI

savings

State value as an outcome with a method. Reserve savings for cash-releasing benefits that finance has agreed.

Adoption

rollout

deployment

enablement

Adoption is people changing how they work; deployment is the technology going live. They are not the same date.

Back to the contents · Suggest a correction (opens GitHub)

Part 10 · Knowledge and navigationChapter 10.2 · 67 of 76

Knowledge and navigation · classification

Taxonomies and ontology

Eight facets to classify anything in the estate, and the relations that connect them. Use the same vocabulary in the register, the ATRS record and the assurance case, or none of the three can be joined up.

Facets

Capability class

Sets which controls carry the weight.

Generative Agentic Declarative

Assurance tier

Derived from criticality, not from popularity.

Personal Team Departmental Operational service Citizen impact

Lifecycle stage

The status field in the register.

Proposed Pilot Live Restricted Suspended Retired

Autonomy level

Guardrails apply at every level.

Suggest Recommend Prepare Act with approval Act and report Act autonomously

Maturity level

Evidence-based, not self-scored.

Uncontrolled Aware Experimenting Governed Integrated Federated Adaptive

Cost layer

Six layers, six owners.

Enterprise foundation Shared platform Product or service Team or workload Use case End user

Evidence strength

Independently validated down to anecdotal.

Strong Moderate Emerging Weak Not assessed

Role family

Determines the learning pathway.

Leadership Product and service Technical and data Governance and assurance Commercial and finance Operations and frontline

Ontology — subject, relation, object

Agent has Purpose, owner, capability class and authority

Agent is recorded in Register entry and ATRS record

Agent is classified by Criticality score, which sets its Assurance tier

Assurance tier requires Evidence set and review frequency

Agent consumes Data sources with provenance and permissions

Agent acts within Autonomy level and guardrails

Agent generates Cost, measured per interaction and per month

Cost is allocated to One of six cost layers with a named owner

Agent affects Service outcome, then public experience, then trust

Incident triggers Failure and recovery loop, then a control change

Owner is accountable for Agent through its whole lifecycle

Evidence supports or refutes Claim, with method, publisher and date

Back to the contents · Suggest a correction (opens GitHub)

Part 10 · Knowledge and navigationChapter 10.3 · 68 of 76

Knowledge and navigation · platforms and repositories

Platforms and repositories

Links point to the vendors' own documentation. Nothing here is an endorsement or a procurement decision: platform choice sits with commercial colleagues, and the dependency and exit chapter applies to every entry on this page.

Government repositories and communities

AI Engineering Lab

Open resource repository under OGL v3.0, plus the Knowledge Hub community

Documentation
i.AI open source

Consult, Redbox, Caddy and the rest of the incubator's public repositories

Documentation
AI Knowledge Hub

Tool pages with cost, evaluation and lessons

Documentation
GOV.UK Design System

Components and patterns for government services

Documentation

Microsoft

Copilot Studio and agent builder

Declarative agents, connectors, publishing controls

Documentation
Microsoft 365 admin and Purview

Permissions, sensitivity labels, audit and retention

Documentation
Azure AI Foundry

Model hosting, evaluation and content safety

Documentation
Power Platform governance

Environment strategy, DLP policies, managed environments

Documentation

Google

Gemini for Workspace admin

Data handling, availability and admin controls

Documentation
Vertex AI

Model garden, evaluation and grounding

Documentation
Google Cloud IAM

Least-privilege permissions and service accounts

Documentation

Atlassian

Atlassian Intelligence and Rovo

Agents over Jira and Confluence content

Documentation
Jira for the pilot backlog

Decision records, findings and remediation tracking

Documentation
Confluence for evidence packs

Session packs, taxonomies and curated patterns

Documentation

Slack

Slack AI and agents

Assistants inside channels, with workspace data boundaries

Documentation
Slack Connect governance

Cross-organisation channels: membership and data controls

Documentation
Workflow Builder

Declarative automation without code

Documentation

Back to the contents · Suggest a correction (opens GitHub)

Part 10 · Knowledge and navigationChapter 10.4 · 69 of 76

Knowledge and navigation · proposed model

From fragmented guidance to a federated service

One trusted discovery layer over many authoritative sources, with a clear path from need to assured deployment. A federated index that links to governed sources beats another copy repository — this is a proposed model synthesised from community discussion and existing guidance, not an authoritative standard.

The landscape today

Government policy & principles
  • AI Playbook for the UK Government
  • Generative AI Framework
  • Civil service guidance
  • Departmental policies
  • Ethical & legal frameworks
Technical documentation
  • Microsoft Learn (Copilot, Agents, Studio, Purview)
  • Technical readiness & configuration
  • Security & identity blueprints
  • Administration guides
Product & licensing information
  • Microsoft 365 Copilot licensing
  • Agent Builder & Studio documentation
  • Capacity, credits & consumption
  • Connectors & capabilities
  • Roadmaps & updates
Local playbooks & templates
  • Departmental playbooks
  • Use case templates
  • Business cases
  • Process maps
  • Agent governance examples
  • Training materials
Community evidence & insight
  • Weekly xGov Copilot calls
  • AI Engineering Lab
  • Community Q&A
  • Worked examples
  • Lessons learned
  • Emerging patterns & issues
Sector & investment guidance
  • FinOps guidance
  • Commercial & procurement guidance
  • Data management & information governance
  • Security, privacy & assurance
  • Accessibility & inclusion
External & market intelligence
  • Supplier blogs & updates
  • Industry reports
  • Standards bodies
  • Academic research
  • International benchmarks

Challenges of fragmentation Hard to find Out of date Inconsistent Duplication Gaps Heavy manual effort Risk of non-compliance

The target state

A federated guidance hub: navigate, connect, understand, apply, assure, share.

Single discovery experience One place to search and navigate.

Authority & evidence clarity Clear source, status and assurance level.

Connected knowledge Linked guidance, tools, templates, use cases and decisions.

Current & relevant Version control, review dates and change alerts.

Role & scenario based pathways Tailored journeys by role, task and maturity level.

Implementation tools Checklists, decision trees, templates, readiness assessments.

Community feedback loop Capture questions, insights and lessons into guidance.

Measure & improve Usage insights, gap analysis, continuous improvement.

Outcomes and benefits

Consistent, proportionate and compliant AI deployment
Faster access to the right guidance and tools
Reduced duplication and wasted effort
Stronger assurance and risk management
Higher quality outcomes and public value

Back to the contents · Suggest a correction (opens GitHub)

Part 11 of 12 · 4 chapters

Deploy for real

Part 11 · Deploy for realChapter 11.1 · 70 of 76

Deploy for real · where the work actually is

Fifty-four public services, placed

The highest-volume services across UK central government, local government and the NHS, placed on two axes that decide almost everything else: how much of it there is, and what happens when it is wrong. Select a quadrant or a factor to filter. Open any service to load it into the generator.

54 services 4 quadrants 10 factors 27 land in Tier 4 34 sit in the dangerous quadrant

One agent, end to end, in each quadrant

Filter by factor

Services

54 of 54 shown

How to read this, and how not to

Volume bands come from published transaction data; criticality and the factor tags are an assessment of consequence, not a classification you can cite. The value is not the placement — it is that placing a service forces a conversation about consequence that a business case usually skips. Move anything you disagree with, and record why you moved it.

Back to the contents · Suggest a correction (opens GitHub)

Part 11 · Deploy for realChapter 11.2 · 71 of 76

Deploy for real · working tool

Derive the governance set for any agent

Two bands and ten factors in. Tier, autonomy cap, oversight regime, currency window, evaluation set, threat priorities, DPIA route, contract clauses, workforce actions, register fields and kill criteria out — with the rules that produced them, so the answer can be argued with rather than accepted.

Volume

Criticality

Extended factors

No factors selected

Q1

Automate with sampling

Tier 1 · Light

The strongest agent case in government. Volume pays for the governance; a mistake is recoverable. The risk here is not harm, it is quietly losing the ability to do the work.

Effective criticality

1 / 14

Autonomy cap

L4

Act with post-hoc sampling — exceptions escalate

Challenge-rate floor

0.5%

Below this, suspect oversight, not excellence.

Sign-off

Service owner

Oversight regime

2% random sample, weekly

Data protection route

Screening assessment, recorded and dated

A screening that concludes no full DPIA is needed is still evidence. Keep it, and re-run it when scope changes.

Evaluations that must exist

Golden-set regression

A frozen set of real historical cases with agreed correct outputs, re-run on every model or prompt change.

Refusal and out-of-scope set

Cases the agent must decline, escalate or flag rather than attempt.

Cost and latency envelope

Per-transaction cost and response time, with the ceiling that triggers review.

Threat priorities

Prompt injection via ingested content High

Any document, email or record the agent reads is untrusted input. Test with adversarial content in every ingestion path.

Tool over-permission High

The agent holds broader credentials than its task needs. Scope per action, not per agent.

Contract clauses that become mandatory

Model change notification

Written notice before any model, weight or default change that affects behaviour, with a defined notice period.

Log and evidence export

All prompts, outputs, tool calls and decisions exportable in a documented open format, during and after the contract.

No training on our data

Explicit prohibition, with the right to evidence of compliance.

Exit and portability

Prompts, instruction blocks, eval sets and grounding indexes are ours, exportable, and documented well enough to rebuild elsewhere.

Workforce actions

Say where the freed capacity went

Volume automation releases hours. Publish where they were redeployed. Unnamed savings become quiet headcount reduction with no decision behind it.

Consultation before design freeze

Staff and unions see the job-redesign proposal while it can still change. After the build is a briefing, not a consultation.

Register fields beyond the baseline

Accountable officer

A named person, not a team or a role inbox.

Reversal path

How an effect is undone, tested, with the last test date.

Currency holders

Who currently holds valid approval authority, and when each window expires.

Kill criteria

Challenge rate falls below floor

Reviewers overturning less than 0.5% of outputs. Either the agent is perfect or oversight has stopped. Assume the second.

Golden-set regression unexplained

Any drop on the frozen set that cannot be explained and fixed within the rollback SLA.

An outcome nobody can explain

A single decision the accountable officer cannot account for in plain language.

Why it produced that answer

  • Low criticality sets a base of 1.
  • Effective criticality 1 of 14 places this in Tier 1 · Light.

Very high volume with low criticality places this in Q1 · Automate with sampling.

The weights are a defensible starting point, not a measured value. Change them with your assurance function and record the change — a tool whose reasoning is hidden cannot be challenged, and a governance answer nobody can challenge is not governance.

Back to the contents · Suggest a correction (opens GitHub)

Part 11 · Deploy for realChapter 11.3 · 72 of 76

Deploy for real · four agents, one per quadrant

Correspondence triage, mandate to operation

Every artefact filled in, including the ones that are uncomfortable. Very high volume, moderate criticality — the case most departments would call easy. Named honestly it lands in Tier 3 · Enhanced, and that is the most useful thing in this chapter.

These four agents are composites, not case studies. The figures are internally consistent and illustrative, and the decisions are the ones the sources support. Your organisation will ratify and adapt every one of them.

Mandate

1 of 10

Signed by the service owner before any build. One page, and it fits.

Problem

Inbound correspondence arrives across six channels into one queue. Median time to first correct routing is four days; 11% is routed wrong at least once. Statutory deadlines are met by escalation, not by process.

In scope

Classify inbound correspondence to one of 34 work queues, extract sender, reference, subject and any stated deadline, and draft a suggested reply from the approved template set.

Explicitly out of scope

Deciding any case. Sending anything. Replying to anyone. Handling correspondence flagged as a complaint, a legal threat, a safeguarding concern or a data-subject request — those route to a human unopened by the agent.

Success measures

Time to correct routing, misroute rate, deadline breaches, and reviewer challenge rate. Not volume processed, and not staff satisfaction alone.

Decision rights

The agent proposes. A caseworker commits. Nothing leaves the organisation without a named person attached to it.

Funding and duration

Twelve months, one delivery team, with a stop point at month four and a re-authorisation decision at month twelve.

What would make us stop

Any statutory deadline missed because of the agent, or a challenge rate below 5% sustained for two months.

What changes the answer inside this quadrant

Same agent, same quadrant, one factor different. Each variant below is scored by the generator on this page, not asserted — so what moves and what does not is the tool's own arithmetic.

The agent sends the reply itself Tier moves

+ Irreversible Tier 4 · Critical Q1 · Automate with sampling L1 · Draft only — a person edits and commits every output

One word in the mandate — send rather than draft — and an effect exists that no later correction undoes. Autonomy is forced to draft-only regardless of tier, and shadow mode becomes mandatory. This is the most common scope change requested after go-live, and the cheapest one to refuse.

Correspondence routinely contains health detail Tier moves

+ Special-category data Tier 4 · Critical Q1 · Automate with sampling L1 · Draft only — a person edits and commits every output

The screening guardrail stops being a nicety and becomes the control the DPIA rests on. A leakage probe joins the evaluation set, and the grounding index has to be provably free of the material.

Internal-only queue, rule-based routing Tier moves

− Discretionary judgement
− Vulnerable users
Tier 2 · Standard
Q1 · Automate with sampling
L3 · Act within guardrails — every action logged and sampled

Remove the discretion and the vulnerable recipient and the same agent, at the same volume, drops a tier. Most of the governance in the worked example is paid for by those two factors, not by the volume.

Back to the contents · Suggest a correction (opens GitHub)

Part 11 · Deploy for realChapter 11.4 · 73 of 76

Deploy for real · scope boundary

What this guide does not cover

A guide that does not state its boundary strands the team at the point where the interesting problems start. This chapter exists so nobody reaches gate 3 believing the architecture, the security testing or the legal advice was somewhere in the earlier parts.

Fully covered here

Principles, criticality and tiering, autonomy boundaries, registers and transparency records, oversight design including currency, cost and value modelling, roles and adoption, the pilot method.

Structured, needs your content

Instruction blocks, evaluation plans, threat profiles, DPIA routes, contract clause sets, job redesign. The shape is here; the substance is yours.

Named but not solved

Reference architecture for your estate, engineering pipeline, executed security testing, model selection, sector regulation, organisational design, safety cases.

What a team must bring

Eight things this guide names but cannot supply. Each one has a normal home in an organisation. Find the home before the build starts, not after.

A target architecture for your estate Enterprise architecture

This guide names what an architecture must satisfy. It does not know your identity provider, your network, your data platform or your existing contracts. Someone with authority over those has to draw it.

Executed security testing Security function or an appointed third party

The threat model here tells you what to test. Running the test — penetration testing, red-teaming, adversarial evaluation — is specialist work with a report at the end of it.

Legal advice on your specific case Legal and your DPO

Everything here about lawful basis, automated decision-making and liability is structure for a conversation with a lawyer. It is not the advice.

Model selection and benchmarking Technical owner

Which model, at what cost, with what evaluation on your data. This guide is deliberately model-agnostic, which means it cannot answer the question you will be asked first.

Sector regulation Your regulator and legal

Health, policing, immigration, education and financial services each carry duties this guide does not enumerate. Where it conflicts with your regulator, your regulator wins.

Organisational design Accountable leadership

Who reports to whom, which team holds the capability, and how it is funded past the pilot. The roles here are role names; you have people, budgets and politics.

Accessibility conformance Accessibility specialist

The comprehension and inclusion tests here are necessary and not sufficient. A conformance audit against WCAG is a separate, evidenced exercise.

Safety cases Clinical or safety governance

Where an agent touches physical safety or clinical outcomes, a structured safety case is required and this guide does not contain one.

Five things to be honest about

Read these before citing anything in this guide to a board, an auditor or a regulator.

Every artefact here is a proposal

Nothing in this guide has statutory force. Until your organisation ratifies it, an artefact is a well-argued suggestion. Ratify it, or expect it to be ignored at the first disagreement.

Figures marked illustrative are illustrative

Where this guide shows a threshold, a percentage or a band, it is a defensible starting point drawn from the sources, not a measured value from your service. Replace it with yours.

Volume and criticality placements are judgements

The fifty-four services in the atlas are placed on published volume bands and an assessment of consequence. They are a starting position for an argument, not a classification you can cite.

The generator does not approve anything

It derives a consistent governance set from the inputs you give it. Give it flattering inputs and it will produce a flattering answer. That is a property of every assessment tool, including the ones with logos on.

The four worked examples are composites

They are built from the sources, from published service data and from patterns that recur across deployments. They are not case studies of four real agents, and no organisation in them is identifiable. Their numbers are internally consistent and illustrative — treat them as a model answer to compare your own against, not as evidence of what happened somewhere.

Sequence is not schedule

The life cycle here is ordered. It carries no durations, because the honest answer depends on your assurance capacity, and that is the constraint nobody plans for.

Back to the contents · Suggest a correction (opens GitHub)

Part 12 of 12 · 3 chapters

Reference

Part 12 · ReferenceChapter 12.1 · 74 of 76

Pan-UK

All four nations have strategy. One has published evaluation.

Counting strategies overstates maturity: a benchmark that measures published strategy ranks all four nations comparable. Only one has a published body of controlled evaluation of deployment. What each nation can borrow from the others is the point of comparing.

UK Government

England and reserved functions

Strategy

AI Playbook (Feb 2025) · AI Opportunities Action Plan (Jan 2025) · departmental AI plans mandated by Cabinet Secretary objectives 2026-27

Who leads

DSIT · Government Digital Service (incl. i.AI and the AI Engineering Lab) · AI Security Institute

Signals

Four published Copilot evaluations plus NAO and PAC scrutiny — the UK's distinctive asset is evaluation, not strategy

ATRS transparency records mandatory for departments and ALBs

Cabinet Secretary's AI Innovation Award (quarterly, from Jun 2026); AI & Data Challenge at 252 ideas

AI Playbook on GOV.UK

Scotland

Devolved strategy, refreshed 2026

Strategy

Scotland's AI Strategy 2026–2031 (March 2026) — five-year plan, OECD-aligned, with the AI Scotland programme

Who leads

Scottish Government · Scottish AI Alliance · The Data Lab

Signals

  • Scottish AI Register gives public-sector use cases a public face
  • Roughly 1 in 3 councils piloting or using AI — leading the UK
  • Independent advisory board plus sector and regional AI Champions
Scotland's AI Strategy

Wales

National AI plan, 2025

Strategy

AI Cymru — the AI Plan for Wales (November 2025); Office for AI established June 2025

Who leads

Welsh Government · Centre for Digital Public Services (CDPS)

Signals

  • Workforce, ethics and service guidance brought together early
  • Welsh-language parity a design requirement for public AI services

Council adoption early-stage (AI Readiness of UK Local Authorities, 2025)

Centre for Digital Public Services

Northern Ireland

Draft strategy, consultation closed 7 October 2026

Strategy

Draft public-sector AI strategy (August 2026) — a draft, not adopted policy; Office of AI and Digital announced 2025

Who leads

The Executive Office · Department of Finance · Office of AI and Digital

Signals

Proposes human-oversight teams, redress mechanisms and AI risk-management standards

Procurement expectations on suppliers: oversight, fairness, transparency, accountability

Stated aim: close the adoption gap with Scotland and Wales

NI Department of Finance

Local-government comparison: AI Readiness of UK Local Authorities 2025 (Heriot-Watt University & GoLLM), 208 councils across all four nations — Scotland leads, with roughly one council in three piloting or using AI.

Back to the contents · Suggest a correction (opens GitHub)

Part 12 · ReferenceChapter 12.2 · 75 of 76

International

Benchmarking the UK's position

Five reference jurisdictions, five different bets: from the EU's binding risk tiers to Singapore's testing-led pragmatism. Genuinely transferable: Canada's public assessment and inventory approach, Australia's named AI ownership and workforce model, Singapore's central adoption flywheel, US inventory discipline, and the EU's joint procurement direction.

Where Key instrument Model For public bodies

UK

AI Playbook + sector regulators; ATRS; AI Security Institute evaluations

Principles-led, pro-innovation; non-statutory for now

Playbook principles and ATRS records; four published evaluations; frontier legislation under consideration

EU

AI Act (in force 1 Aug 2024; Digital Omnibus 2026 defers high-risk duties to Dec 2027, public authorities to Aug 2030); EuropAI joint procurement

Binding, risk-based regulation

High-risk public deployments: conformity assessment, logging, human oversight, fundamental-rights impact assessment

US

NIST AI Risk Management Framework; OMB memoranda; GAO oversight

Voluntary framework + agency rules; state laws vary

Federal AI use-case inventories and Chief AI Officers; rapid growth with persistent policy and budget challenges

Canada

Directive on Automated Decision-Making + Algorithmic Impact Assessment; agentic-AI guidance (May 2026)

Binding on federal automated decisions

Mandatory AIA before deployment; names skill atrophy as an oversight risk; bounded autonomy and recoverability

Singapore

Model AI Governance Framework · AI Verify toolkit

Voluntary, testing-led

Central funding, shared platforms, problem intake, applied learning — an adoption flywheel

The EU AI Act's four risk tiers

In force 1 August 2024. The Digital Omnibus on AI (Regulation (EU) 2026/1744, in force 27 July 2026) moved stand-alone high-risk obligations to 2 December 2027, embedded high-risk systems to 2 August 2028 and high-risk systems for public authorities to 2 August 2030. Article 50 transparency duties have applied since 2 August 2026. Obligations follow the use case, not the technology. The benchmark other regimes are measured against.

Unacceptable

Banned outright: social scoring, subliminal manipulation, real-time remote biometric ID in public spaces (narrow exceptions).

High risk

Essential public services, law enforcement, employment, critical infrastructure: conformity assessment, documentation, human oversight, EU database registration.

Limited risk

Transparency duties: chatbots must say they are AI; synthetic media must be labelled.

Minimal risk

Most systems: spam filters, recommenders. No mandatory obligations; voluntary codes encouraged.

The supranational layer

OECD AI Principles + OECD.AI Observatory

The shared definition of AI used by the UK Playbook and the EU AI Act. The OECD.AI Policy Observatory tracks national AI policies worldwide; it is the model for this app's observatory.

Council of Europe Framework Convention on AI (2024)

The first binding international AI treaty: human rights, democracy and the rule of law. The UK is a signatory.

G7 Hiroshima AI Process (2023)

International guiding principles and a code of conduct for organisations developing advanced AI systems.

AI Safety Summit series: Bletchley, Seoul, Paris

Frontier-safety declarations and the network of national AI safety and security institutes. Started at Bletchley Park, November 2023.

UN General Assembly resolution (2024)

First UN consensus resolution on safe, secure and trustworthy AI for sustainable development.

Back to the contents · Suggest a correction (opens GitHub)

Part 12 · ReferenceChapter 12.3 · 76 of 76

Reference · pace and parity

Named once. Controlled nowhere.

Skill atrophy appears in this guide as a single sentence in the evidence chapter, credited to Canada. Nothing sits behind it: no control, no metric, no register field, no gate criterion, no owner. That is the honest answer to the question, and it is now backlog item PACE-01. This chapter closes it, then applies the same test to everything else the world has built that we have not.

Named

Yes

One sentence, evidence chapter, attributed to Canada

Controlled

No

No control, no owner, no frequency

Measured

No

No metric in the barometer or control tower

Gated

No

No criterion in the quality gate

What Canada actually did

Six moves, and the last is the operable one. The instrument is the Guide on the Use of Agentic Artificial Intelligence, published by the Treasury Board of Canada Secretariat on 22 May 2026, sitting on top of a directive that already made assessment mandatory.

Named the risk inside an instrument

Continued offloading of decisions to AI agents can lead to knowledge and skill atrophy and weakened human judgment. Employees become less practiced at core tasks and slower to notice when something is wrong.

Guide on the Use of Agentic AI, TBS, 22 May 2026

Tied it to the validity of oversight

Maintenance of relevant skill sets among agentic AI users and deployers is treated as a prerequisite for meaningful human oversight. Not a wellbeing concern — a validity condition on the control itself.

Same instrument, mitigations

Added two agent-specific principles

Bounded autonomy and recoverability, introduced alongside the existing generative-AI principles rather than replacing them.

Same instrument

Named the consequence chain

Service quality, privacy and security, legal defensibility and public trust. The risk is framed as legal and operational exposure, not a training preference.

Same instrument

Made assessment mandatory and public

Algorithmic Impact Assessment before deployment, four impact levels, external peer review at levels III and IV, questionnaire published openly including on GitHub.

Directive on Automated Decision-Making

Pointed at a concrete drill

Periodic human-only execution of agent-handled tasks for comparison; staff trained to interrogate agent outputs rather than accept them.

Commentary on the guide, May 2026

PACE-01 · proposed instrument

Oversight currency

Canada names skill maintenance as a prerequisite for meaningful human oversight. A prerequisite that nobody measures is a sentiment. Aviation, nuclear operations and clinical practice all solved this the same way: recency requirements, evidenced, with authority attached. If you supervise the work, you stay current in the work. Six components, all of them auditable.

Component What it does Who holds it Evidence

Manual-mode drill

A reviewer executes an agent-handled task unaided, then compares. Tier 1 monthly, tier 2 quarterly, tier 3 twice yearly.

Reviewer

Drill record holding both outputs

Currency window

A reviewer whose last unaided execution predates the tier window loses approval authority on that queue until re-current. Enforced, not advised.

Service owner

Authority state in the register

Divergence log

Every drill records agreement or disagreement and why. Sustained perfect agreement is itself a signal — it is what rubber-stamping looks like from outside.

Reviewer

Divergence entries with reasons

Challenge rate

Proportion of agent outputs materially amended before acceptance. A falling challenge rate against a flat error rate is drift, not improvement.

Control tower

Monthly trend per queue

Rotation

No reviewer supervises the same agent-handled queue continuously. Rotation preserves unaided practice somewhere in the role.

Line manager

Rota showing unaided work

Displaced-skill field

The register records which skills the agent displaces, who must retain them and to what standard. If nobody is named, the deployment is not approvable.

Deployment owner

Register field, tested at each gate

The uncomfortable component is the divergence log. A reviewer who agrees with the agent every time is either supervising a perfect agent or not supervising at all, and the record cannot tell the difference. Convergence has to be read as a warning rather than a result.

Eight instruments elsewhere, measured against ours

Three are outright gaps: something binding or published elsewhere with no UK counterpart. Four are partial. One is context we should track rather than copy.

EU AI Act Article 4 — the AI literacy duty Gap

Binding since 2 February 2025 and unaffected by the proposed high-risk delay. Applies to deployers of every AI system, not only high-risk ones, and covers staff and contractors regardless of seniority. The compliance test is a dated, role-mapped programme that plausibly matches actual use.

Where we stand

Training is encouraged throughout UK guidance but is nowhere a duty, and course completion is not evidence of competence.

Singapore Model AI Governance Framework for Agentic AI Gap

World-first agentic framework, Davos 22 January 2026; version 1.5 on 20 May 2026 after feedback from over sixty organisations, with more than ten deployment case studies. Four pillars, unique agent identities tied to a supervising agent or user, threat modelling for memory poisoning, tool misuse and privilege compromise.

Where we stand

We hold secure-AI guidance and our own patterns, but no single agentic instrument and no agent identity requirement.

Canada

Directive on Automated Decision-Making, plus the agentic guide

Gap

Assessment is mandatory before deployment, scored across four impact levels, with external peer review at the top two. The questionnaire is published openly, including on GitHub, and has become the instrument others copy.

Where we stand

ATRS is transparency after the fact, not assessment before it, and it is not a gate.

Australia AI Safety Institute inaugural report, 9 August 2026 Open to all

Found that NIST's risk framework, OWASP's Agentic Top 10, Singapore's agentic framework and ASD's agentic guidance all rest on the same hidden assumption: that one organisation owns every agent in a system. Cross-organisational agent risk is a structural gap none of them covers.

Where we stand

Our register assumes single-owner agents too. The gap is identified and unimplemented everywhere, including here — which makes it available.

Global Singapore Consensus companion report on agentic risk Partial

Ten foundational principles for managing agentic risk: least privilege, traceable identity, auditability, validated deployment, adversarial resilience, multi-agent stability, runtime assurance, interruptibility, legibility and human oversight.

Where we stand

We hold most of these as scattered controls. None is consolidated into a named principle set an approver can test a deployment against.

OECD Digital Government Outlook 2026 Partial

Thirty-two of thirty-six countries run AI training programmes; only thirteen of thirty-six train on using AI in public services or in policymaking. Broad capability is improving faster than role-specific readiness. Thirty of thirty-six have an institution governing public-sector AI; twenty-one of thirty-six offer central procurement support.

Where we stand

Our academy is role-based on paper. Depth of oversight-specific training has never been tested against this benchmark.

G7 GAIN network and the GovAI Grand Challenge Partial

A G7 AI network convened and chaired by Canada, tasked with a roadmap to scale successful public-sector AI projects and a catalogue of open-source, shareable AI solutions for members.

Where we stand

We participate. We have contributed no reusable instrument to the catalogue.

Asia-Pacific Korea AI Basic Act (2026), Taiwan AI Basic Act (2025) Tracking

Regional convergence on human oversight, data safeguards and prohibitions on harmful or deceptive use, arriving through basic law rather than sectoral guidance.

Where we stand

Context rather than gap: the UK route is guidance-led by choice. Worth tracking for interoperability, not imitation.

The backlog, MoSCoW'd

Thirteen items and three deliberate refusals. Every item has an owner and a first deliverable, because an unowned backlog item is a wish.

Must

This quarter. Each one closes a gap that leaves an existing control unsound.

PACE-01
Oversight currency

Manual-mode drills, currency windows, divergence log, rotation and the displaced-skill register field. Turns the Canada finding into something an auditor can test.

Owner: Learning lead with service owners First: Control spec and tier windows

PACE-02
Evidenced literacy duty

A dated, role-mapped competence requirement for anyone who operates or approves an agent, contractors in scope. The Article 4 test, applied voluntarily and ahead of any obligation.

Owner: Learning lead First: Role-to-competence map

PACE-03
Agent identity

Every agent carries a unique identity bound to a named human owner and, where orchestrated, to a supervising agent. Prerequisite for audit, for handover and for retirement.

Owner: Architecture First: Identity schema in the register

PACE-04
Bounded autonomy and recoverability

Adopt both as named approval principles. Every agent has a tested interrupt and a tested rollback before it goes live, with the test evidenced.

Owner: Assurance lead First: Two gate criteria and a test script

PACE-05
Drift telemetry

Challenge rate and divergence surfaced in the control tower beside assurance status, per queue, monthly.

Owner: Control tower First: Two new tiles

Should

Next two quarters. These extend reach rather than repair a fault.

PACE-06
Cross-boundary agents

Register fields for agents we interact with but do not own: supplier, other department, arm's-length body. The first implementation of the gap Australia named.

Owner: Architecture with commercial First: Field set and one worked case

PACE-07
Multi-agent systemic risk

An assessment step for orchestrated patterns, covering interaction effects that no single-agent assessment can see.

Owner: Assurance lead First: Assessment addendum

PACE-08
Oversight-specific training

Not general AI awareness. Training in interrogating outputs, recognising plausible-but-wrong, and running a manual-mode drill.

Owner: Academy First: Curriculum for three roles

PACE-09
Assessment before deployment

A scored pre-deployment assessment distinct from the transparency record, with independent peer review at the top tiers.

Owner: Governance office First: Scoring instrument

PACE-10
Extended threat model

Memory poisoning, tool misuse and privilege compromise added to the agent threat model and to the security review.

Owner: Security lead First: Threat model revision

Could

Worth doing, sequenced behind the above. This is where the leadership case is actually made.

PACE-11
Publish the instruments

Register schema, gate criteria and the currency control released openly under the Open Government Licence, versioned, as Canada did with its assessment.

Owner: Governance office First: Public repository

PACE-12
Reciprocal assurance pilot

One partner nation, one deployment class, mutual recognition of assurance evidence. Tests whether our evidence travels before we offer it widely.

Owner: Governance office First: Pilot terms with one partner

PACE-13
Atrophy measurement study

Longitudinal measurement of unaided performance in agent-supported roles, with an academic partner. Nobody holds this data, which is why everyone is asserting rather than measuring.

Owner: Evaluation lead First: Study protocol

Won't — for now

Named, so the omission reads as a decision rather than an oversight.

PACE-W1
Statutory literacy penalties

Not ours to set. We build the evidence base a policy decision would need.

Owner: — First: Revisit when policy opens

PACE-W2
A UK agentic act

The guidance route moves faster, and we do not yet hold the evidence to legislate well.

Owner: — First: Revisit after PACE-13

PACE-W3
Certifying supplier agents

Capacity we do not have. Contract terms and evidence requirements instead.

Owner: — First: Revisit at scale

Four things nobody has published yet

Parity is a floor, and reaching it puts us level with instruments already a year old. Each of these is a control we could be first to publish, and three of the four came out of our own case work rather than the international scan.

Oversight currency as a revocable authority

Canada names skill maintenance as a prerequisite for oversight. Nobody has made it measurable and revocable — an approval right you can lose by going stale. Publishing the control with tier windows and drill records would be first.

The orphaned-agent control

Machinery-of-government change moves an owner; active agent work continues unattended and is later inherited by a team that never approved it. No framework surveyed handles ownership transfer on reorganisation.

Attention budgets for review

Approval capacity treated as a finite, costed resource and rationed by risk tier, so review is real where it matters instead of uniform and hollow everywhere.

Cross-boundary agent handshake

Australia's institute named cross-organisational agent risk as a gap no framework covers. A register field plus a handshake at departmental and supplier boundaries would be the first implementation of it.

Keeping pace is a rhythm, not a document

Four of the eight instruments above were published or revised in the last eight months. A guide refreshed annually is a guide that is wrong for eleven of them.

Weekly

Horizon scan into the guidance ticker: new instrument, new version, new enforcement action.

Scan rota

Fortnightly

Instrument diff across the eight tracked jurisdictions. What changed, what it obliges, whether it touches a control we hold.

Governance office

Monthly

Re-MoSCoW the backlog. Anything sitting in Should for two cycles is promoted or killed, never carried.

Governance office

Quarterly

Publish the delta pack and contribute one reusable instrument to a shared catalogue.

Governance office

Annually

Reciprocal review with a partner nation: they read our instruments, we read theirs, both publish the findings.

SRO

Where the work goes once it exists

Five routes already open to us. None of them takes a position paper; all of them take a working instrument.

G7 GAIN

The network's remit includes a catalogue of open-source, shareable AI solutions for members. A register schema and a currency control are exactly that shape.

OECD.AI and the Digital Government Outlook

The comparative baseline every country is measured against. Contributing method, rather than only returning survey answers, changes what the baseline measures.

Council of Europe Framework Convention

The first binding international AI treaty, on human rights, democracy and the rule of law. The UK is a signatory, and the validity of human oversight sits squarely inside its scope.

The AI safety institute network

The UK allocated roughly A$460m at the 2025 spending review, against about A$50m over five years in Canada and about A$11m a year in Singapore. Capacity to convene is not the constraint.

Open publication

Open Government Licence v3.0, public repository, versioned. Canada published its assessment questionnaire openly and it became the reference others adopted.

Sources: Guide on the Use of Agentic Artificial Intelligence and Directive on Automated Decision-Making, Treasury Board of Canada Secretariat; EU AI Act Article 4 and the AI Office Q&A on AI literacy; IMDA Model AI Governance Framework for Agentic AI v1.5 (20 May 2026); Australian AI Safety Institute inaugural report (9 August 2026); Singapore Consensus companion report on agentic risk; OECD Digital Government Outlook 2026 and Building an AI-ready public workforce (2026); G7 GovAI Grand Challenge and GAIN, Treasury Board of Canada Secretariat and Impact Canada. Figures are as published by each source and are not converted or combined.

Back to the contents · Suggest a correction (opens GitHub)

Working tool

Can I use AI for this?

Five questions give a first route. Governance follows what the AI can see, create and do, not which company made it. Nothing you enter leaves your browser.

1 Would the output decide, or directly change, someone’s rights, entitlements, safety or access to a service?
2 Will you use personal data or information marked OFFICIAL-SENSITIVE or above?
3 Can the AI take actions, run on a schedule or work through several steps without you?
4 Is there an approved tool in your organisation's catalogue that already does this?
5 Could a conventional rule, form or automation do the job reliably?

Answer the five questions

The route appears when every question has an answer. “Not sure” is a valid answer.

How the route is decided

The first rule that matches wins. A first sort, not an approval: every route other than an approved tool leads to a registered use case.

  1. Would the output decide, or directly change, someone’s rights, entitlements, safety or access to a service?
  2. Will you use personal data or information marked OFFICIAL-SENSITIVE or above?
  3. Can the AI take actions, run on a schedule or work through several steps without you?
  4. Is there an approved tool in your organisation's catalogue that already does this?
  5. Could a conventional rule, form or automation do the job reliably?
WhenRouteWhat it means
Q1 is YesSeek advice before going furtherAI must not make or directly change this kind of decision on its own. Register a use case; it takes the specialist route with a named human decision-maker.
Q5 is YesTry conventional automation firstA rule, form or workflow is usually safer, cheaper and more consistent. Use AI only where judgement over unstructured material is the task.
Q3 is Yes or Not sureTake the agent routeAnything that acts, runs on a schedule or works through steps without you needs an agent design record, approval points and a stop control.
Q1 is Not sure, or Q2 is Yes or Not sureRegister a use casePersonal or sensitive information, or an unclear impact on people, needs data readiness and privacy screening before use.
Q4 is YesGo ahead with an approved toolUse it within its approved uses and check its catalogue entry for conditions. Review every output before you rely on it.
OtherwiseCheck for reuse, then registerNo approved tool is known to fit. Search your organisation's catalogue first; if nothing fits, request a tool through a use case.

Knowledge map

How the playbook fits together

Parts on the inner ring, chapters on the outer ring; select any node to go to it. The taxonomy is published as SKOS in JSON-LD and the content as JSON, so other portals can harvest both.

Knowledge map of the playbook12 parts on an inner ring, each linked to its chapters on the outer ring. The same structure is listed as text in the contents drawer.PlaybookOverview (Start here)Can I use AI for this? (Start here)How to use this guide (Start here)The playbook (Start here)Capability domains (Start here)1Part 1: Start hereEvidence before assertion (Decide and prioritise)Use cases & evidence (Decide and prioritise)Public sector case library (Decide and prioritise)Classify & assess (Decide and prioritise)Criticality & gates (Decide and prioritise)Assurance tiers (Decide and prioritise)2Part 2: Decide and prioritiseLifecycle & templates (Design and build)Capability classes (Design and build)Skill studio (Design and build)Agent assurance X-ray (Design and build)Connectors, MCP & integrations (Design and build)Architecture & quality (Design and build)Knowledge foundation (Design and build)Autonomy boundaries (Design and build)3Part 3: Design and buildControl plane & evidence (Assure and govern)Data & information governance (Assure and govern)Architecture of trust (Assure and govern)Governance operating model (Assure and govern)Risks & mitigations (Assure and govern)Registers & lifecycle (Assure and govern)Internal ATRS registry (Assure and govern)Failure & recovery (Assure and govern)4Part 4: Assure and governPortfolio control tower (Operate and oversee)Monitoring & reassessment (Operate and oversee)Agent journey (Operate and oversee)Agent employment (Operate and oversee)Public service impact (Operate and oversee)Observatory & barometer (Operate and oversee)5Part 5: Operate and overseeCost, value & investment (Fund and buy)FinOps & models (Fund and buy)Commercial & suppliers (Fund and buy)Dependency & exit (Fund and buy)6Part 6: Fund and buyMultidisciplinary teams (People and adoption)AI service team (People and adoption)Roles & pathways (People and adoption)Capability academy (People and adoption)Capability & maturity (People and adoption)Inclusive adoption (People and adoption)Adoption journey (People and adoption)Agent rollout (People and adoption)Community engagement (People and adoption)Operating model (People and adoption)Adoption & change model (People and adoption)7Part 7: People and adoptionThe pilot in one view (Live pilot)What it tests (Live pilot)Candidate selection (Live pilot)Pilot roles (Live pilot)Assurance model (Live pilot)Journey & ceremonies (Live pilot)Communications (Live pilot)Success & feedback (Live pilot)Decision & toolkit (Live pilot)8Part 8: Live pilotProduct requirements (Product documentation)Functional & technical spec (Product documentation)Solution architecture (Product documentation)High-level design (Product documentation)Low-level design (Product documentation)Theory of change (Product documentation)Personas & journeys (Product documentation)Gov.uk AI resources (Product documentation)9Part 9: Product documentationThesaurus (Knowledge and navigation)Taxonomies & ontology (Knowledge and navigation)Platforms & repositories (Knowledge and navigation)Federated guidance service (Knowledge and navigation)10Part 10: Knowledge and navigationService atlas (Deploy for real)Case generator (Deploy for real)Worked examples (Deploy for real)What this does not cover (Deploy for real)11Part 11: Deploy for realUK nations (Reference)International (Reference)Pace & parity (Reference)12Part 12: Reference
  1. 1. Start here 5
  2. 2. Decide and prioritise 6
  3. 3. Design and build 8
  4. 4. Assure and govern 8
  5. 5. Operate and oversee 6
  6. 6. Fund and buy 4
  7. 7. People and adoption 11
  8. 8. Live pilot 9
  9. 9. Product documentation 8
  10. 10. Knowledge and navigation 4
  11. 11. Deploy for real 4
  12. 12. Reference 3

Taxonomy (SKOS, JSON-LD) · Content (JSON) · Glossary

Transparency

How the playbook was made

Ported from an internal AI agent governance toolkit on 18 November 2026. Each chapter was rendered, classified as public or withheld, redacted where needed and rebuilt as plain semantic content. Figures the source labels illustrative stay labelled illustrative; nothing here is an official policy position. The source has 87 chapters: 76 are published here and 11 are withheld or merged into the rest of the portal. The decision for every chapter is in the content classification (opens GitHub).

Withheld or merged chapters

ChapterPartDecision and reason
AI tool catalogueDecide and prioritiseOrganisation-specific approved-tool statuses and links to internal pages; publishing them would read as endorsement.
Analytics, estimators & dashboardsDecide and prioritiseIts charts count entries in the withheld tool catalogue; the estimators it links to are published in "Cost, value and investment".
Blueprint coverageProduct documentationCoverage map of internal source documents and blueprints, which are not published.
Portal specification traceProduct documentationTraces requirements from an internal specification document.
Knowledge graphKnowledge and navigationInteractive graph of the internal edition; replaced by the generated knowledge map in this playbook.
Portal & site mapKnowledge and navigationMap of the internal edition, including internal pages. The public playbook has its own contents and drawer.
Guided toursKnowledge and navigationGuided tours route through internal pages; the home page personas and journeys replace them.
Media centreKnowledge and navigationFilms and decks: never kept in a repository (owner rule). They stay in the media library outside GitHub.
Self-assurance consoleKnowledge and navigationSelf-test console of the internal edition; the public site runs its own checks in CI (see Transparency).
Sources & blueprintsReferenceLists internal source documents. Public sources are in the resource library on the home page.
GlossaryKnowledge and navigationMerged into the portal glossary, so the whole portal has one glossary: 39 new terms added, existing definitions kept.

How changes are made and tracked

  • Reviewed every quarter; next review 2027-02-18.
  • The playbook, the contents drawer, the hero figures and the masthead menus are generated from data/playbook.json by tools/build_playbook.py; check S21 fails the build if they drift.
  • Check S22 fails the build if any published file names an internal page, an internal team or an email address.
  • Corrections and requests go through GitHub issue forms (opens GitHub); the roadmap lists what is next.
Working tools

Do the work here, not in a separate spreadsheet

Six tools that turn the guidance into an artefact you can paste into a ticket, a change record or a board paper. Everything runs in your browser — nothing you type is sent anywhere, stored or transmitted.

Privacy by design: these tools are client-side only. Do not enter personal data, case details or anything above OFFICIAL. Use references and descriptions instead.

1. Human review triage — what level of oversight does this need?
About the output
Does any of the following apply?

Your result will appear here.

2. AI-use declaration and evidence pack generator

Produces a copy-ready record. Attach it to the artefact, the ticket or the change record.

Be specific: what did AI do, and what did you do?
For an answer an agent built from a shared glossary or semantic layer: term, the column it maps to, and its review level. Leave empty if none.

Your result will appear here.

3. DEIA impact screen — equality, accessibility and inclusion

A screening tool, not a substitute for a full equality impact assessment. It tells you whether you need one.

Tick everything that could apply

Your result will appear here.

4. Transparency record generator (ATRS-shaped)

Drafts a public-facing record of how an algorithmic or AI tool is used. Publish it, and link it from the service.

Your result will appear here.

5. Quantum readiness and crypto-agility assessor
Tick what is true today

Your result will appear here.

6. Plain-language and inclusive content checker

Paste AI-drafted content to check reading difficulty, sentence length, jargon and non-inclusive phrasing before publication.

Nothing leaves your browser. Do not paste personal data.

Your result will appear here.

Toolkit

A complete operating kit for AI-assisted public delivery

Every asset is designed to be catalogued, versioned, reviewed and reused. Filter by purpose, or search across names, descriptions and linked standards.

Searches asset names, descriptions and linked standards.

Showing 18 of 18 toolkit assets.

TK-04AssuranceSmall effortLive tool

Evidence pack builder

Source consultation records, AI-use declarations, validation results, human acknowledgements and audit trails.

Output: Generator on this page

Anchored to: ATRS (opens an external website)

Open the tool

TK-07AssuranceSmall effortLive tool

Human review triage (H0–H5)

Risk-based routing that decides how many eyes, which specialists and what evidence an AI-assisted output needs.

Output: Triage tool on this page

Anchored to: NIST AI RMF (opens an external website)

Open the tool

TK-10GovernanceSmall effortLive tool

Transparency record generator

Produce an ATRS-shaped public record for any algorithmic or AI-assisted tool in use.

Output: Generator on this page

Anchored to: ATRS hub (opens an external website)

Open the tool

TK-17GovernanceMedium effort

Records, retention and FOI readiness

Classify AI interaction logs as records; set retention, disclosure and redaction rules ahead of FOI/SAR requests.

Output: Retention schedule

Anchored to: DPA 2018 (opens an external website)

TK-18DeliveryMedium effort

Capability and competence pathway

Role-based learning paths, licence-to-operate checks and refreshers for anyone using AI on public work.

Output: Curriculum map

Anchored to: DDaT framework (opens an external website)

Assurance

Human-centred controls for trustworthy AI work

Risk-based human-in-the-loop controls make sure AI never grants authority a user does not already possess, and that every material output can be defended after the fact.

The five-stage assurance workflow

  1. Classify — identify service impact, data sensitivity, equality and accessibility risk, security risk and model/agent autonomy. Use the triage tool →
  2. Select — apply the standards, source guidance, skills, agents, conformance packs and review level the classification demands.
  3. Execute — work with least privilege, mandatory source citation, deterministic checks and pair or buddy support where competence is thin.
  4. Review — 4-eyes or 6-eyes review, specialist assurance, and no self-approval for material or high-risk outputs.
  5. Learn — capture evidence, incidents, near-misses, feedback, source drift and model changes; feed them back into the corpus.

The H0–H5 human review model

Review levels, when they apply and the evidence each produces
LevelNameWhat it meansTypical examplesEvidence
H0No review neededPersonal, throwaway, non-published exploration with no personal or sensitive data.Explaining a concept to yourself; drafting private notes.None required.
H1Self-checkAuthor verifies facts, sources and accessibility before use. Low impact, easily reversible.Internal meeting notes, first-draft ideas.AI-use note in the document.
H2Peer checkA second competent person checks output before it leaves the team.Internal reports, non-public analysis, low-risk code in a sandbox.Reviewer name, date, issues found.
H3Specialist review (4-eyes)A named specialist in the relevant discipline reviews. No self-approval.Accessibility, security, legal, statistical or clinical content; production code.Specialist sign-off + test results.
H4Multi-specialist review (6-eyes)Two or more specialisms plus the accountable owner. Independent challenge required.Public-facing decisions, casework support, anything affecting entitlements.Panel record, dissent captured, residual risk stated.
H5Governance board and publicationFormal board approval, published transparency record and monitored deployment.Automated or semi-automated decisions about people; national services.ATRS record, DPIA, EQIA, monitoring plan, review date.

Reviewer checklist

Always check

  • Every citation opens, exists and says what the text claims
  • Numbers, dates, thresholds and legal references independently verified
  • Nothing asserts compliance that has not actually been tested
  • No personal or sensitive data was entered into an unapproved tool
  • Accessibility and plain-language requirements met, not assumed
  • Assumptions, gaps and residual risk stated explicitly
  • Reviewer is independent of the author at H3 and above

Common failure patterns

  • • Confident fabrication — plausible citations that do not exist
  • • Stale authority — quoting withdrawn or superseded guidance
  • • Compliance theatre — a checklist ticked without a test run
  • • Silent scope creep — an agent doing more than was authorised
  • • Automation bias — reviewers rubber-stamping fluent output
  • • Averaged users — outputs tuned to a majority that excludes minorities
  • • Evidence gaps — no record of who checked what, or when

Accountability at a glance (RACI)

Who does what in AI-assisted delivery
ActivityResponsibleAccountableConsultedInformed
Classifying risk and review levelAuthorService ownerSpecialist reviewersDelivery team
Verifying sources and outputsAuthorAuthorPeer reviewerService owner
Specialist sign-off (H3+)Named specialistService ownerAuthorAssurance board
Equality and accessibility screeningAuthorService ownerEquality & accessibility leadsGovernance board
Data protection and lawful basisProduct ownerData controller / SIRODPOICO where required
Publishing a transparency recordService ownerSROComms, legalPublic
Incident and near-miss reportingAnyone who spots itService ownerSecurity, DPOAssurance board
Metadata catalogue

Treat skills, agents and code as assets in their own right

Every reusable AI capability should be discoverable, accountable, interoperable and governed like any other public asset — with an owner, a version, a risk rating and a retirement plan.

Catalogue fields

Minimum viable metadata for any AI asset
FieldWhy it matters
asset_idStable identifier so evidence, incidents and dependencies can point at it.
ownerA named person, not a team inbox. Accountability needs a face.
statusdraft / approved / restricted / deprecated / withdrawn.
versionSemantic versioning; breaking changes force re-review.
authority_tier0–3 tier of the sources the asset relies on.
risk_level / human_reviewH0–H5 floor that consumers must apply.
source_provenanceWhich guidance, data and code it derives from, with dates.
permissionsLeast-privilege scopes, tools and credentials the asset may use.
prohibited_actionsExplicit red lines an agent must never cross.
deia_screenDate and outcome of equality and accessibility screening.
evaluationTest suite, pass thresholds and last run.
compatibilityModels, runtimes and versions it is validated against.
lifecycleReview cadence, next review, deprecation and retirement date.
contact / redressHow to report a problem or challenge an outcome.

Example skill asset metadata

asset_type: ai_skill
asset_id: review-accessibility-govuk
title: GOV.UK accessibility review assistant
status: approved
version: 2.1.0
owner: AI assurance lead (role, not a named person)
authority_tier: 1
human_review: H3
specialist_review: [accessibility, content_design]
self_approval: false
source_provenance:
  - id: wcag22
    url: https://www.w3.org/TR/WCAG22/
    checked: 2026-07-14
    next_check: 2026-10-14
  - id: govuk-design-system
    url: https://design-system.service.gov.uk/
    checked: 2026-07-14
permissions:
  tools: [read_page, run_axe]
  data_classification_max: OFFICIAL
prohibited_actions:
  - claiming_conformance_without_test_evidence
  - processing_personal_data
  - autonomous_publication
deia_screen:
  completed: 2026-06-02
  outcome: mitigations_applied
evaluation:
  suite: a11y-regression-v4
  threshold: 0.95
  last_run: 2026-08-01
compatibility:
  models: [">=v3.1"]
  runtimes: [node20, python3.12]
evidence_rules:
  require_citations: true
  require_reviewer_signoff: true
  retain_for: P7Y
lifecycle:
  review_cadence: P3M
  next_review: 2026-11-01
  deprecation: null
contact: https://github.com/cyberdudeuk/ai-capability-governance/issues/new

Interoperability note: map these fields to DCAT 3 (opens an external website) for cataloguing, PROV-O (opens an external website) for evidence chains and CycloneDX ML-BOM (opens an external website) for supply chain, so registries in different organisations can federate rather than fork.

UCD, DEIA and accessibility

Designed for real people, varied needs and accountable teams

Inclusion is a delivery activity, not a compliance report. This section states what we commit to, how we test it, and how you can hold us to it.

Design and research practice

Write for reading age 9

Plain English, short sentences, no unexplained acronyms, inclusive terminology, and alternative formats on request.

Check your content →

Accessibility commitments for this portal

What we build in

  • Semantic HTML, landmarks and a logical heading order
  • Full keyboard operation with a visible, high-contrast focus indicator
  • Text contrast of at least 4.5:1 and 3:1 for interface components
  • Light, dark and high-contrast themes, plus a larger-text setting that persists
  • Respect for prefers-reduced-motion and no motion-only meaning
  • Responsive to 320 CSS px and to 400% zoom without loss of content
  • Status messages announced through ARIA live regions
  • No colour-only meaning; icons always paired with text
  • Descriptive link text that makes sense out of context
  • A print stylesheet that expands links and collapsible content

How we test and report

  • Automated checks in the pipeline on every change
  • Manual keyboard and screen reader testing before release
  • Periodic testing with disabled participants
  • A published accessibility statement listing known gaps and fix dates
  • A named contact and a 5-working-day response commitment
  • Escalation route to the Equality Advisory and Support Service

Accessibility statement guidance → (opens an external website)

DEIA in AI specifically

Known equity risks of AI-assisted delivery. Performance that degrades for under-represented groups; language models that default to majority dialects and names; outputs written above the reading age of the audience; automation that removes the human route people in crisis rely on; accessibility regressions in AI-generated code and content; and reviewers who trust fluent text more than accurate text.

What we do about them. Screen every change with the DEIA tool; disaggregate evaluation by affected group where lawful and proportionate; keep a human route that cannot be removed by efficiency targets; test AI-generated interfaces with assistive technology before merge; publish limitations honestly; and give people a real way to challenge an outcome.

Code of practice and conduct

Responsible use of AI assistive technology

This code complements behavioural codes of conduct and the Civil Service values. It defines operational rules for model-backed tools, direct AI use, coding assistants, multi-agent orchestration and AI-generated artefacts.

Core duties
  • Stay accountable. You own every output you use, send or publish. "The AI wrote it" is never a defence.
  • Disclose material assistance. If AI shaped the substance, structure, analysis or code, say so.
  • Verify before you rely. Open every citation. Recalculate every figure. Test every claim of compliance.
  • Protect data. Only approved tools, only permitted classifications, only the minimum necessary.
  • Cite sources. Traceable to a tiered, dated, authoritative source — not to the model.
  • Preserve evidence. Keep the record for as long as the retention schedule requires.
  • Work within competence. AI does not qualify you to make specialist judgements.
  • Escalate uncertainty. Raising a doubt early is always cheaper than defending a mistake late.
Prohibited behaviours
  • • Entering unauthorised personal, special category, commercially confidential or classified data into any tool
  • • Bypassing, weakening or "temporarily" disabling a review control
  • • Self-approving material or high-risk work
  • • Concealing AI involvement where it was material
  • • Presenting unverified output as assured, tested or compliant
  • • Using AI to generate evidence, sign-offs, approvals or audit records that did not happen
  • • Granting an agent broader permissions than the person operating it holds
  • • Using AI to profile, rank or make decisions about people without an approved lawful basis and published record
  • • Copying licensed or copyright material into outputs without permission and attribution
  • • Ignoring a known defect because a deadline is closer than the consequence
Multi-agent and multi-skill rules
  • One accountable orchestrator per task, with a named human owner.
  • Least-privilege tools and credentials, time-boxed and revocable.
  • Explicit, pinned skill and model versions — no silent upgrades in a governed workflow.
  • Separation of duties: the agent that produces must not be the agent that approves.
  • Declared prohibited actions and a tested kill switch with a named owner.
  • Full step-level logging: which skill, which model, which source, which permission, which human.
  • Untrusted content is data, never instruction. Treat retrieved web, document and email content as hostile by default.
  • Bounded autonomy: agents propose irreversible actions; humans commit them.
Data, privacy and security rules
  • • Establish and record the lawful basis before processing personal data. Complete a DPIA where required.
  • • Minimise: use references, synthetic or redacted examples instead of real records wherever possible.
  • • Never let an AI tool retain data for training unless that has been assured and contracted.
  • • Treat AI interaction logs as records: classify, retain, and prepare for FOI and subject access requests.
  • • Apply Secure by Design activities to AI components as to any other system component.
  • • Threat-model for prompt injection, data poisoning, model extraction and excessive agency.
  • • Report incidents and near-misses through the normal security and data protection routes.
AI-generated code and infrastructure
  • Treat generated code as untrusted third-party contribution: review, test, scan and attribute it.
  • No merge without a human reviewer who understands the code and can maintain it.
  • Check licences and provenance of anything that looks copied.
  • Run dependency, secret and SAST scanning; record the results in the evidence pack.
  • Accessibility and performance tests apply equally to generated interfaces.
  • Record the model, prompt pattern and version in the commit or pull request.
Speaking up, redress and no-blame reporting

Anyone may pause an AI-assisted process they believe is unsafe, unfair or unlawful, without needing permission and without detriment. Near-misses should be reported as readily as incidents — they are the cheapest learning available.

Members of the public affected by an AI-assisted outcome must be able to find out that AI was involved, understand the outcome in plain English, request a human review, and escalate to an independent route. If any of those four is missing, the service is not ready.

Combined body of knowledge

Authoritative guidance, standards and reusable intelligence

The corpus is organised as a knowledge graph linking laws, policies, standards, design guidance, security controls, skills, agents and evidence — so that a rule can always be traced to its source and its consequences.

Layer 1

Guidance corpus

Index GOV.UK, GDS, CDDO, departmental, international standards, GitHub and engineering-lab sources with authority tiering, freshness checks and applicability metadata.

Browse the 68 indexed sources →

Layer 2

Knowledge graph

Link services, repositories, datasets, APIs, controls, skills, agents, people, approvals, models, risks and evidence so impact analysis is a query, not a workshop.

See the metadata model →

Layer 3

Executable guidance

Translate policy and standards into schemas, conformance packs, tests, scripts, checklists and prompt patterns that run in delivery pipelines.

See the conformance packs →

Authority tiering — how conflicts are resolved

When two sources disagree, the lower tier number wins. Record the conflict and the resolution in the evidence pack.

Source authority tiers used across the corpus
TierSource typeExamplesTreatment
0Law and statutory instrumentsEquality Act 2010, DPA 2018, Accessibility Regulations 2018Binding. Non-compliance is not a risk to accept — escalate.
1Mandated standards and cross-government policyService Standard, Technology Code of Practice, Secure by Design, WCAG 2.2, ATRSComply or record a formal, time-boxed exception with an owner.
2Recognised international and professional standardsISO/IEC 42001, NIST AI RMF, OWASP LLM Top 10Adopt as good practice; justify divergence.
3Community, vendor and research materialBlogs, model cards, toolkits, preprintsUseful signal. Never the sole basis for a control or a public claim.

Freshness and drift controls

Every indexed source carries

  • • Authority tier and issuing body
  • • Date checked and next check due
  • • Applicability (organisation, service type, data classification)
  • • Supersession chain and withdrawal status
  • • Linked controls, skills and conformance tests

Drift triggers a review when

  • • A cited source is updated, withdrawn or superseded
  • • A model or its version changes materially
  • • A control fails in production or a conformance test breaks
  • • An incident, complaint or FOI request touches the artefact
  • • The scheduled review date passes without confirmation
Resource library

Guidance, standards, toolkits and solutions — indexed and tiered

68 curated sources, each with its authority tier and a note on what it is actually useful for. Search, filter, then follow the link to the source of truth.

Searches titles, publishers, topics and descriptions.

Showing 69 of 69 resources.

Indexed guidance, standards and tools. Tier 0 is law; tier 3 is community or vendor material.
ResourcePublisherTypeTierUse it for
AI Playbook for the UK Government (opens an external website)
governancedeliveryuk
DSITPolicyTier 1Ten principles plus per-technology guidance for safe, effective public sector AI adoption. Replaced the Generative AI Framework for HMG, withdrawn on 10 February 2025.
Algorithmic Transparency Recording Standard (ATRS) (opens an external website)
transparencygovernanceuk
DSIT / CDDOStandardTier 1Mandatory-style template for publishing how an algorithmic tool is used in decision-making.
Data Ethics Framework (opens an external website)
ethicsdatauk
DSITPolicyTier 1Structured ethical appraisal for public sector data and AI projects.
Service Standard (14 points) (opens an external website)
ucddeliveryuk
GDSStandardTier 1The assessment backbone for any public service — align AI features to each point.
Government Service Manual (opens an external website)
ucddeliveryuk
GDSGuidanceTier 1End-to-end practice guidance for user research, design, technology and operations.
Government Design Principles (opens an external website)
ucduk
GDSGuidanceTier 1Ten principles — start with user needs, do less, design with data, do the hard work to make it simple.
Technology Code of Practice (opens an external website)
deliverygovernanceuk
CDDOStandardTier 1Criteria for technology spend approval — interoperability, open standards, security, sustainability.
Digital, Data and Technology Capability Framework (opens an external website)
skillspeopleuk
GDSToolkitTier 1Map AI-related skills to recognised roles and progression levels.
Open source guidance (coding in the open) (opens an external website)
engineeringopennessuk
GDSGuidanceTier 1When code should be open or closed, and how to publish safely.
GDS API technical and data standards (opens an external website)
engineeringinteroperabilityuk
GDSStandardTier 1How to design, document and version APIs that agents and skills can call.
WCAG 2.2 (opens an external website)
accessibilitydeia
W3CStandardTier 1The normative success criteria — target AA as a minimum, AAA where feasible.
ARIA Authoring Practices Guide (APG) (opens an external website)
accessibilityengineering
W3C WAIGuidanceTier 1Accessible patterns for tabs, accordions, comboboxes, dialogs and more.
EN 301 549 accessibility requirements (opens an external website)
accessibilityprocurement
ETSI/CEN/CENELECStandardTier 1The procurement-facing accessibility standard referenced by UK/EU regulations.
Accessibility requirements for public sector websites and apps (opens an external website)
accessibilitylegaluk
GOV.UKGuidanceTier 1What the 2018 regulations require, including the accessibility statement duty.
Public Sector Bodies (Websites and Mobile Applications) Accessibility Regulations 2018 (opens an external website)
accessibilitylegaluk
legislation.gov.ukLawTier 0The underlying legal duty.
Equality Act 2010 (opens an external website)
deialegaluk
legislation.gov.ukLawTier 0Protected characteristics, reasonable adjustments and discrimination duties.
Public Sector Equality Duty guidance (opens an external website)
deialegaluk
EHRCGuidanceTier 1Due regard duty — evidence it in every AI-assisted decision or service change.
Make your service accessible: an introduction (opens an external website)
accessibilityucduk
GDSGuidanceTier 1Practical starting point for teams new to accessibility duties.
Publish an accessibility statement (opens an external website)
accessibilitytransparencyuk
GOV.UKToolkitTier 1Model wording and required disclosures for non-accessible content.
Research with people with disabilities and access needs (opens an external website)
researchdeiauk
GDSGuidanceTier 1How to recruit and run inclusive sessions, including with assistive technology users.
Helping people to use your service (assisted digital) (opens an external website)
deiainclusionuk
GDSGuidanceTier 1Assisted digital support and digital inclusion planning.
Inclusive Design Principles (opens an external website)
ucddeia
Paciello/TPGi et al.GuidanceTier 2Seven principles: provide comparable experience, be consistent, give control, offer choice.
Inclusive Design Toolkit (opens an external website)
ucddeia
MicrosoftToolkitTier 3Persona spectrum and exclusion-mapping activities for workshops.
Home Office accessibility "dos and don’ts" posters (opens an external website)
accessibilitydeiauk
UK Home OfficeToolkitTier 1Print-ready posters for autistic users, screen readers, dyslexia, low vision, D/deaf users.
GOV.UK Design System (opens an external website)
ucdengineeringuk
GDSToolkitTier 1Tested, accessible components and patterns — reuse before you invent.
GOV.UK Prototype Kit (opens an external website)
ucdresearchuk
GDSToolkitTier 1Rapid, realistic prototypes for usability testing without production risk.
GOV.UK style guide (content design) (opens an external website)
contentdeiauk
GDSGuidanceTier 1Plain English, reading age, formatting and inclusive terminology.
WebAIM contrast checker (opens an external website)
accessibilitytool
WebAIMToolTier 3Verify 4.5:1 text and 3:1 non-text contrast before shipping.
Pa11y automated accessibility testing (opens an external website)
accessibilitycitool
Pa11yToolTier 3Add automated WCAG checks to your pipeline — never as a substitute for manual testing.
axe DevTools (opens an external website)
accessibilitytool
DequeToolTier 3Browser-based rule engine used widely for triage.
Guidelines for secure AI system development (opens an external website)
securityaiuk
NCSC + international partnersGuidanceTier 1Secure design, development, deployment and operation across the AI lifecycle.
Secure by Design (UK government) (opens an external website)
securitydeliveryuk
Cabinet Office / GSGStandardTier 1Mandated activities and artefacts for digital service security in HMG.
Cyber Assessment Framework (CAF) (opens an external website)
securityassuranceuk
NCSCStandardTier 1Outcome-based objectives A–D for essential services and government systems.
Post-quantum cryptography guidance and migration timeline (opens an external website)
pqcsecurityuk
NCSCGuidanceTier 1Discovery by 2028, high-priority migration by 2031, complete by 2035.
Post-Quantum Cryptography standards (FIPS 203/204/205) (opens an external website)
pqcsecurity
NISTStandardTier 1ML-KEM, ML-DSA and SLH-DSA — the algorithms your crypto inventory should target.
OWASP Top 10 for LLM Applications (opens an external website)
securityaiengineering
OWASPStandardTier 2Prompt injection, insecure output handling, supply chain and agency risks.
MITRE ATLAS adversarial ML knowledge base (opens an external website)
securityaithreat
MITREToolkitTier 2Tactics and techniques for threat modelling AI systems.
Secure Software Development Framework (SP 800-218) (opens an external website)
engineeringsecurity
NISTStandardTier 2Baseline secure SDLC practices to bind AI-generated code to.
SLSA supply chain integrity framework (opens an external website)
engineeringsupplychain
OpenSSFStandardTier 2Provenance levels for build artefacts — extend to model and skill artefacts.
CycloneDX ML-BOM / AI-BOM (opens an external website)
metadatasupplychain
OWASP CycloneDXStandardTier 2Machine-readable bill of materials for models, datasets and AI components.
Guidance on AI and data protection (opens an external website)
privacylegaluk
ICOGuidanceTier 1Lawful basis, fairness, transparency and accountability for AI processing. Under review by the ICO after the Data (Use and Access) Act 2025.
AI and data protection risk toolkit (opens an external website)
privacyassuranceuk
ICOToolkitTier 1Risk statements and controls mapped to the AI lifecycle. Under review by the ICO after the Data (Use and Access) Act 2025.
Data protection impact assessments (DPIA) (opens an external website)
privacylegaluk
ICOToolkitTier 1When a DPIA is mandatory and what a defensible one contains. Under review by the ICO after the Data (Use and Access) Act 2025.
Data Protection Act 2018 (opens an external website)
privacylegaluk
legislation.gov.ukLawTier 0UK GDPR implementation. Its automated decision-making rules were rewritten by the Data (Use and Access) Act 2025.
Data (Use and Access) Act 2025 (opens an external website)
privacylegaluk
legislation.gov.ukLawTier 0Replaced UK GDPR Article 22 with Articles 22A to 22D from 5 February 2026: a significant decision taken with no meaningful human involvement needs specific safeguards, including the right to contest it.
Automated decision-making, including profiling (opens an external website)
privacyassuranceuk
ICOGuidanceTier 1What counts as meaningful human involvement, and the safeguards significant automated decisions need. Draft updated for the 2025 Act; the consultation closed on 29 May 2026.
ISO/IEC 42001 AI management systems (opens an external website)
governancecertification
ISO/IECStandardTier 1Certifiable management system for responsible AI — useful for supplier assurance.
ISO/IEC 23894 AI risk management (opens an external website)
riskgovernance
ISO/IECStandardTier 1AI-specific application of ISO 31000 risk management.
ISO/IEC 27001 information security management (opens an external website)
securitycertification
ISO/IECStandardTier 1Control baseline most suppliers will already hold.
AI Risk Management Framework (AI RMF 1.0) + GenAI Profile (opens an external website)
riskgovernance
NISTStandardTier 1Govern, Map, Measure, Manage — pairs cleanly with UK assurance practice.
EU AI Act (Regulation 2024/1689) (opens an external website)
legalriskeu
EUR-LexLawTier 0Risk tiers and obligations that may bind UK bodies operating in the EU market.
OECD AI Principles (opens an external website)
ethicsinternational
OECDPolicyTier 2Internationally agreed values-based principles for trustworthy AI.
Recommendation on the Ethics of AI (opens an external website)
ethicsinternational
UNESCOPolicyTier 2Global normative instrument covering human rights and inclusion.
Understanding AI ethics and safety (opens an external website)
ethicsukresearch
The Alan Turing InstituteGuidanceTier 2The SUM values and FAST Track principles used across UK public sector.
AI Standards Hub (opens an external website)
standardsuk
BSI / NPL / TuringToolkitTier 2Searchable database of AI standards in development and published.
Ada Lovelace Institute research (opens an external website)
ethicspublicresearch
Ada Lovelace InstituteResearchTier 3Public attitudes, algorithmic impact assessment and deliberative methods.
Open Data Institute (opens an external website)
dataopenness
ODIResearchTier 3Data ethics canvas and data institution patterns.
Model Cards for Model Reporting (opens an external website)
metadatatransparency
Mitchell et al.ResearchTier 2The reference format for documenting intended use and performance by group.
Datasheets for Datasets (opens an external website)
metadatadata
Gebru et al.ResearchTier 2Provenance, composition and collection questions for training/eval data.
PROV-O provenance ontology (opens an external website)
metadataprovenance
W3CStandardTier 1Entity–activity–agent model for machine-readable evidence chains.
DCAT 3 data catalogue vocabulary (opens an external website)
metadatainteroperability
W3CStandardTier 1Describe datasets and services so registries can federate.
C2PA content provenance and authenticity (opens an external website)
provenancegenai
C2PAStandardTier 2Cryptographic manifests for AI-generated or edited media.
Model Context Protocol (opens an external website)
agentsengineeringinteroperability
MCPStandardTier 3Open protocol for connecting assistants to tools and data with explicit scopes.
Model cards on Hugging Face Hub (opens an external website)
metadataengineering
Hugging FaceToolkitTier 3Practical templates and YAML front-matter conventions.
alphagov on GitHub (opens an external website)
engineeringopennessuk
GDSToolkitTier 1Reference implementations, frontend, and publishing platform source.
x-govuk community tools (opens an external website)
engineeringcommunityuk
Cross-government communityToolkitTier 2Community-built plugins, design tooling and prototypes.
Open Government Licence v3.0 (opens an external website)
opennesslegaluk
The National ArchivesLawTier 0Default licence for publishing public sector content and guidance.
Greening Government ICT and Digital Services Strategy (opens an external website)
sustainabilityuk
Defra / GGCPolicyTier 1Carbon, circular-economy and reporting expectations for digital estates. The 2020 to 2025 period has ended; the 2024 to 2025 annual report (July 2026) says a new government digital sustainability strategy is coming.
Web Sustainability Guidelines (WSG) (opens an external website)
sustainabilityengineering
W3C Community GroupGuidanceTier 3Practical efficiency criteria for pages, media, hosting and AI workloads.

Content reviewed on 2026-11-13: every link checked, automatically or by hand in a browser where a site refuses automated checks, and each page read for withdrawal, replacement or review notices. Next review due 2027-02-13. Guidance changes without notice: always confirm you are reading the current version at the source.

Horizon scan · AI and data governance

What practitioners and regulators are saying

29 signals on governing AI agents and the data they use, gathered from public LinkedIn content and refreshed weekly. Each names its author, links to where it was found and says how far it has been checked. Inclusion is not endorsement.

Read with care. LinkedIn does not allow automated reading of posts, so every summary comes from search-indexed text and is paraphrased. 15 of 29 come from vendors, consultancies or law firms and are marked. Statistics without a primary source are flagged as unverified. Check the source before you quote anything. The register is open data: ai-data-governance-signals.json (scanned 2026-11-09, version 1.2.0).

S01UKUK public sectorData foundationsCommercial interest

UK public sector data leaders on data in action for public services

Lee Stockwell, Databricks · 2026

Activity highlights George Bettsworth, Head of Data Science at the Foreign, Commonwealth and Development Office, putting AI progress down to good governance, with the argument that governance should sit at the data layer rather than in each application. Also trails an Innovation 2026 panel with Sue Bateman (Defra CDO), Jacqui Leggetter (DWP) and Aimee Smith (Government Chief Data Officer).

Why it matters: Direct UK government voices on governing AI through the data estate; supports BOT's position that agent governance starts with data governance.

Caution: Vendor-hosted content; the FCDO view is relayed by the vendor, not quoted from FCDO.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S02UKAssurance & standardsUK public sectorAgent governance

Trusted third-party AI assurance: from principles to implementation

Elizabeth Seger, Tony Blair Institute for Global Change · 2026 (about one month before scan)

Argues AI governance is shifting from principles to implementation and that governance capacity is what lets organisations adopt AI agents safely. Points to the UK Government's Trusted Third-Party AI Assurance Roadmap (September 2025), tied to commitment 29 of the AI Opportunities Action Plan.

Why it matters: Links BOT's assurance pages to a named UK policy instrument and commitment number.

Caution: Roadmap and commitment number should be checked against GOV.UK before citing in BOT's evidence base.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S03UKAgent governanceUK public sector

Scaling the Responsible Adoption of Agentic AI (techUK report)

Usman Ikhlaq, techUK (author); now University of Bath Institute for Policy Research · 21 July 2026

Profile lists a techUK report he authored on moving agentic AI beyond pilots to responsible deployment at scale across the UK economy, including how oversight and accountability keep pace.

Why it matters: UK industry baseline for agent oversight; a comparator for BOT's agent governance playbook.

Caution: Report itself not read; summary is from the profile listing.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S04UKAgent governanceAssurance & standardsUK public sector

AI Security Institute and NCSC advice on agentic AI cyber risk

AI Security Institute (AISI), UK Government · 2026

AISI's page links to NCSC guidance on managing the cyber risk of agentic AI and to its Alignment Red Team work. The page describes AISI as part of the Cabinet Office.

Why it matters: Authoritative UK security sources BOT should cite for agent permissions and runtime controls.

Caution: Departmental home is as stated on the LinkedIn page; confirm on aisi.gov.uk given recent machinery of government changes.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S05UKAgent governanceRegulation & policyUK public sectorCommercial interest

Agent autonomy, accountability and control boundaries in financial services

Keith Vincent (speaker); hosted by Ashurst Perkins Coie, Government Legal Department, Treasury Legal Advisers · 7 October 2026

Event activity frames AI agents around autonomy, accountability and where control boundaries sit, and notes that customers' own agents bring new liquidity and fraud risks.

Why it matters: Government legal perspective on the same question BOT answers: what an agent may do without a human.

Caution: Event summary written by the host firm, not by the speaker.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S06UKRegulation & policyAgent governanceCommercial interest

FCA moving towards agentic supervision

Sophie Sheldon (quoted in Law360 UK); liked by Devina Shah, Simmons & Simmons · 2026

Activity reports the FCA's plan to become an agentic regulator, with a warning about over-reliance on a single model, platform or provider, and separately references new FCA guidance on AI in financial services.

Why it matters: A UK regulator itself adopting agents: a live case for BOT's concentration-risk and vendor-neutrality guidance.

Caution: Secondary report via Law360; check the FCA's own statement.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S07UKUK public sectorData foundationsAssurance & standards

Responsible AI governance: beyond the 'Department of No'

Yasmin Liverpool, Chief Data Office, National Highways · Undated (likely 2025)

LinkedIn article proposing shared accountability for AI governance (for example the CDO with the CSO) or a cross-executive AI steering group, citing the NAO finding that poor data quality is a barrier to AI in government and the Generative AI Framework for HM Government as a starting point.

Why it matters: A practising government CDO's operating-model view; maps onto BOT's roles and accountability pages.

Caution: Predates later UK guidance that replaced the Generative AI Framework; treat as historical context.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S08UKAgent governanceCommercial interest

Most organisations will adopt agents before they can govern them

Melvyn White, Agentic AI strategy, governance and management (independent, London) · Current profile

Profile thesis: the gap between agent adoption and the ability to govern agents is the core risk.

Why it matters: A crisp framing of the governance-debt problem BOT exists to close.

Caution: Opinion; no evidence attached in the indexed text.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S09InternationalData foundationsRegulation & policy

Data governance must scale with AI adoption

Jake Boudsocq, Department of Government Services (Victoria, Australia) · 2026 (about five months before scan)

Takeaways from the 2026 Melbourne Data Protection and Security Summit: data governance is the operating model that makes responsible AI possible at scale; uncatalogued legacy data compounds risk; Digital Transformation Agency model clauses are becoming a public sector baseline; AI governance is not set-and-forget.

Why it matters: Peer-government evidence for BOT's 'know your data before AI' gate and for catalogue-first sequencing.

Caution: Australian context; DTA clauses are not UK instruments.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S10InternationalAgent governanceAssurance & standards

2026 AI governance priorities, including governing AI agents

Partnership on AI (reposted by Stephanie Ifayemi), Partnership on AI · 2026 (about eight months before scan)

Sets out priorities led by governing AI agents with accountability, security and privacy, plus stronger documentation and public reporting across the AI value chain and shared international baselines. Also notes work on how public sector AI governance differs from private sector approaches.

Why it matters: Multi-stakeholder agenda that BOT's agent registry and transparency pages already address.

Caution: Agenda-setting piece; not guidance.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S11InternationalAgent governanceAssurance & standardsCommercial interest

From governance decisions to enforceable authority at runtime

Patrick Upmann, AIGN Global · 2026

As agents move from generating information to taking actions across data, tools and systems, the question becomes how governance decisions are enforced at runtime. Situates this against ISO/IEC 42001, the NIST AI Risk Management Framework and the EU AI Act.

Why it matters: Matches BOT's 'the app is the leash' principle: permissions enforced by scope, not policy text.

Caution: Author markets a proprietary governance framework.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S12InternationalAgent registries & metadataAgent governanceCommercial interest

Agent scanners and an agent registry built on agent-card metadata

Roberto Mandiola, Amazon Web Services (post about MuleSoft and Google Cloud) · 2026

Describes scanners that find AI agents and capture metadata on what each can do, its underlying model and its data access rights, mapped to the agent-to-agent protocol's agent card so organisations hold one registry of agents, including home-built agents and MCP servers.

Why it matters: Strongest metadata link in this scan: an agent register is a metadata problem, which ties BOT to the Metadata Exchange Model and DCAT work.

Caution: Vendor announcement relayed in another person's activity feed (indexed under Andrea Radojicic's profile).

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S13InternationalData foundationsAgent governanceCommercial interest

Governing AI agents as data consumers

Bigeye (and others in the feed), Bigeye · 2026

Argues that governing agents that read enterprise data differs from human-era data governance because agents fail at different points in the pipeline; recommends knowing data quality signals and what data agents access before rollout. The feed also cites a Sinch survey of 2,527 senior decision-makers in which 74% had rolled back a live agent after a governance failure.

Why it matters: Useful framing for BOT's data-access controls; the rollback figure is a candidate indicator.

Caution: Survey figure is vendor-reported and unverified; do not use in BOT's evidence base until the Sinch source is read.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S14InternationalAgent governanceAgent registries & metadataCommercial interest

Unsanctioned agents as governance debt

Subeer Sehgal, Grazitti Interactive · 2026

A playbook for governing agents: where risk concentrates, which controls hold and how to prove compliance as agent numbers grow. Warns that unvalidated feedback loops, where agents treat their own inferences as fact, are the real blind spot. The company feed claims 54% of organisations have between 1 and 100 unsanctioned agents.

Why it matters: Supports BOT's agent inventory and lineage controls.

Caution: The 54% figure has no stated source; treat as unverified.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S15InternationalData foundationsAgent registries & metadata

Context engines and a knowledge layer for agents

Robert Noyce, Shell · 2026

Agents need business context, semantic definitions, governance, memory and trusted data; a knowledge layer linking business definitions to data gives agents that context.

Why it matters: Endorses the glossary, ontology and knowledge-graph layer the govukOS build standard already requires.

Caution: Practitioner reflection after an event; no evidence attached.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S16InternationalData foundationsCommercial interest

AI-ready data goes beyond data quality

Joyce Goh, Salesforce (Informatica) · 2026

Data readiness affects AI economics, and governance must evolve as AI moves from generating insight to agents taking action, needing more visibility, accountability and control.

Why it matters: Corroborates S01 and S09 from the platform side.

Caution: Vendor promotional content.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S17InternationalData foundationsCommercial interest

Agents stretch the old data quality problem

Herman Yu, Teeup Data Technology · 2026

Data governance and data quality issues are long-standing; the non-deterministic behaviour of LLMs and agents stretches them further, so enterprise architecture must join governance dimensions into one framework.

Why it matters: Supports treating agent governance as an extension of data governance, not a new silo.

Caution: Opinion.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S18EURegulation & policyCommercial interest

EU AI Act in force but enforcement readiness uneven

Robert Gasparovic, HAVEL & PARTNERS, Slovak Republic · 2026

Notes the EU AI Act has applied since 2 August 2026 but Member States differ in enforcement readiness; Europe is also revisiting text-and-data-mining rules.

Why it matters: Cross-border context for BOT's regulatory crosswalk.

Caution: Applicability dates differ by obligation under the Act; check the specific article before citing.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S19InternationalRegulation & policyAgent governanceCommercial interest

China's July 2026 framework for governing AI agents

Pascal Jiang (post by Rimon PC), Dacheng Law Offices / Rimon PC · July 2026

Three measures effective July 2026 move China from principles to operational governance, including AI ethics guidance, a framework for anthropomorphic AI services and a policy framework for governing AI agents.

Why it matters: One of the first national-level agent governance frameworks; a comparator for BOT.

Caution: Law-firm summary; the primary texts are not read.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S20InternationalUK public sectorAssurance & standards

Institutional readiness for responsible AI in the public sector

Anthony Akpan, Tizee Consult; Institutional AI Governance newsletter · April and July 2026

An SSRN preprint (April 2026) on institutional readiness as a condition for responsible AI in the public sector, and a newsletter launched in July 2026 on institutional responsibility in AI governance.

Why it matters: Academic framing of readiness that maps onto BOT's maturity and adoption model.

Caution: Preprint, not peer reviewed; focus is developing countries.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S21InternationalAssurance & standardsData foundationsCommercial interest

Five AI governance platforms compared

Brad Martineau, LinkedIn article · Undated

Compares Credo AI, IBM watsonx.governance, Holistic AI, Fiddler AI and Atlan, noting that a data governance and metadata platform such as Atlan supplies the lineage regulators expect.

Why it matters: Market scan only; useful for BOT's tooling landscape, not for recommendations.

Caution: Listicle; possible commercial interest. BOT stays vendor-neutral.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S22UKUK public sectorAssurance & standards

AI Security Institute publishes its inaugural Frontier AI Trends Report

Tess Buckley, Post about the AI Security Institute (DSIT) · 18 December 2025 (derived from the post's activity ID)

Post announcing that the UK AI Security Institute, part of DSIT, published its first Frontier AI Trends Report.

Why it matters: A UK government evidence source on frontier AI capability and risk that BOT's assurance material can cite once read at source.

Caution: Only the headline and URL were indexed; LinkedIn refuses page fetches, so the body was not read. Read the report on GOV.UK before citing any finding. Older than the last scan date; indexed only now.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S23InternationalData foundationsRegulation & policy

AI governance as the next frontier of data governance

Khyati Thakkar, LinkedIn article · Undated

Article framing AI governance as an extension of existing data governance practice.

Why it matters: Supports BOT's premise that AI and agent governance should be built on data governance foundations.

Caution: Only the headline and URL were indexed; LinkedIn refuses page fetches, so the body was not read. Author name taken from the article URL; organisation and date unknown.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S24InternationalAssurance & standards

Developing AI assurance frameworks

Heather Dawe, LinkedIn article · Undated

Article on how organisations can build AI assurance frameworks.

Why it matters: Possible practitioner input to BOT's assurance pillar.

Caution: Only the headline and URL were indexed; LinkedIn refuses page fetches, so the body was not read. Author name taken from the article URL; organisation, region and date not confirmed.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S25InternationalAgent governanceAgent registries & metadata

AI control planes for governing models, agents and decisions at scale

Mahmoud Abufadda, LinkedIn article · Undated

Article on the rise of AI control planes as a way to govern models, agents and their decisions centrally.

Why it matters: Relates to BOT's control-tower and agent-registry thinking: one governed layer over many agents.

Caution: Only the headline and URL were indexed; LinkedIn refuses page fetches, so the body was not read. Author name taken from the article URL; organisation and date unknown. Check for vendor interest before citing.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S26UKUK public sectorAssurance & standards

UK introduction to AI assurance

Katharina Koerner, LinkedIn post · 15 February 2024 (derived from the post's activity ID)

Post sharing the UK government's Introduction to AI assurance guidance.

Why it matters: Points to the baseline UK assurance guidance BOT should align with.

Caution: Only the headline and URL were indexed; LinkedIn refuses page fetches, so the body was not read. Dated 2024, so background rather than new signal; check GOV.UK for the current edition.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S27UKUK public sectorAssurance & standards

CETaS paper on assurance of third-party AI systems for UK national security

The Alan Turing Institute (CETaS), The Alan Turing Institute · 17 January 2024 (derived from the post's activity ID)

Turing Institute post on a CETaS paper about AI risk management and assuring third-party AI systems in a UK national security setting.

Why it matters: Complements S02 on third-party assurance with a public-sector research source.

Caution: Only the headline and URL were indexed; LinkedIn refuses page fetches, so the body was not read. Full paper title is truncated in the URL; dated 2024, so background rather than new signal.

Source on LinkedIn (opens an external website) · evidence: search-indexed text, post not opened

S28InternationalAgent registries & metadataData foundationsCommercial interest

Semantic layer for agents: AI drafts business definitions, people certify them, agents answer in SQL

NVIDIA, Auto Ontology (open source, Apache 2.0) · November 2026 (repository read on 13 November 2026)

An open-source service that drafts an ontology from database metadata: business terms mapped to columns, reviewed SQL expressions, and the joins between tables. People then test it with real questions and certify the names, descriptions and SQL. An agent answers questions by writing SQL from those definitions, checks the query against the database, and returns the answer with its SQL and the steps it took. It serves this context to other agents over MCP, and says when a question cannot be answered from the data.

Why it matters: It makes certification a separate human step from drafting, which is where this toolkit's H0 to H5 review levels apply: see the advanced guide's worked examples on certifying a definition and reviewing an agent's answer, and the evidence pack's field for the definitions an answer relied on.

Caution: Vendor project, and it does not accept outside contributions. Accuracy is measured with a separate text-to-SQL benchmark repository; this review did not run it or the software.

Source repository on GitHub (opens an external website) · evidence: primary source, the repository's README (summary and how it works)

S00InternationalRegulation & policyAssurance & standardsAgent governanceCommercial interest

AI governance as four questions: the law, the proof, the accountability, and what runs in production

Luís Rodrigues, Not stated on the post (the author's headline describes him as a CPTO; he writes the newsletter Build What Matters) · About 7 November 2026 (the post showed '1w' when read on 17 November; the date is decoded from its activity ID)

Frames AI governance as four layers: regulation (the EU AI Act, plus GDPR, sector rules, NIS2 and DORA), standards and frameworks (ISO/IEC 42001 and the NIST AI RMF), roles (a named owner for every AI system) and controls. Argues that controls are where most organisations fall short, and lists them: risk-rate every system, give agents only the tools they need, train people in AI literacy, keep a human able to stop the system, keep an inventory, log everything, and test continuously. Suggests starting with an inventory of every AI tool in use.

Why it matters: The four questions map onto this toolkit: the law (Resources, tier 0), proof (evidence pack builder, TK-04), accountability (the accountable owner and H4 to H5 review) and production (agent registry, TK-03, and incident reporting, TK-12).

Caution: A practitioner's opinion post that also promotes the author's newsletter. The legal points (for example AI literacy as an EU duty) and the standards named should be checked against the primary texts before citing.

Source on LinkedIn (opens an external website) · evidence: post read in full in the owner's signed-in browser on 17 November 2026

Credit belongs to the named authors. To correct or remove an item, use the contact routes below.

Reference

Glossary and frequently asked questions

Glossary

4-eyes / 6-eyes
Two or three independent people involved in producing and approving an output.
Agent
Software that plans and acts across multiple steps using tools, within explicit permissions and bounds.
Agentic
A capability that takes actions across tools under delegated authority, choosing its own path to a goal.
Assisted digital
Support for people who cannot use a digital service independently.
Assurance tier
The proportionate level of control an agent attracts, derived from criticality rather than popularity.
ATRS
Algorithmic Transparency Recording Standard — the UK format for publishing algorithmic tool use.
Authority
What an agent is permitted to do, as distinct from what it is technically capable of doing.
Automation bias
The tendency to accept an automated output without enough scrutiny.
Autonomy level
How far an agent may act without a human decision, from suggest-only to bounded execution.
Barometer
The eight-dimension readiness self-assessment held on the reader's own device.
Capability class
Whether a deployment is generative, agentic or declarative — the determinant of which controls carry the weight.
Conformance pack
A bundle of automated checks that tests an artefact against a named standard.
Connector
An integration that lets an AI tool read from, or write to, another system.
Criticality
Operational dependency and consequence of failure, scored across ten dimensions to set the gate.
Crypto-agility
The ability to change cryptographic algorithms without redesigning the system.
Data boundary
A supplier commitment about where data is stored and processed. Check which features and models sit outside it.
Declarative
A capability defined by a specification — instructions, sources, permitted actions and constraints — that the platform executes.
Declarative agent
An agent defined by configuration: purpose, instructions, knowledge and permitted actions, run on a platform’s own model and orchestration.
DEIA
Diversity, equity, inclusion and accessibility.
Deployment route
The same model reached directly, through a cloud platform or inside another supplier’s product is a different governance object each time.
Deprecation
A supplier’s notice that a model, feature or interface will be withdrawn.
Deterministic automation
Rules-based automation that gives the same output for the same input every time.
Digital worker
An agent framed as a managed member of the workforce, with recruitment, probation, review and retirement.
DPIA
Data protection impact assessment, required for high-risk processing.
Drift
Divergence between a cited source, a model version or a control and what is actually in use.
EQIA
Equality impact assessment evidencing the public sector equality duty.
Evidence card
The standard record of a claim: what was done, method, result, publisher, date and known limitations.
Evidence pack
The bundle of records proving how an AI-assisted output was produced, checked and approved.
Exit path
The seven-step route out of a platform, model or supplier dependency: discover, target, design, extract, test, cutover, decommission.
FinOps
The practice of joining consumption, cost, quality and value so spend is forecast, allocated and optimised.
Generative
A capability that produces content, analysis or options; nothing changes until a person acts on the output.
Grounding
Supplying authoritative source material so outputs can be traced, cited and checked.
Guardrail
A constraint that holds regardless of instruction: prohibited actions, data limits, escalation triggers.
Harvest now, decrypt later
Capturing encrypted data today to decrypt once quantum computers are capable.
Human in the loop
A person who must act before an output takes effect.
Human on the loop
A person who monitors and can intervene in an otherwise automated process.
Human-agent system
The agent, the people who instruct and review it, and the controls between them, tested as one.
Information asset owner
The named person accountable for an information asset, its use, sharing and risks.
Instruction block
The seven-part specification of an agent's behaviour: goal, context, sources, task, format, guardrails, output.
Investment line
The full cost of adoption: visible, foundational, people and change, operational, and transition and exit.
Kill switch
A control that stops an agent immediately and blocks further actions.
Least privilege
Granting only the minimum permissions needed for the shortest necessary time.
Lock-in
Dependency on a platform, model, connector or supplier that raises the cost or risk of leaving.
Material AI assistance
AI involvement that a reasonable reader would want disclosed because it shaped the output.
Meaningful human involvement
Under the UK GDPR, a significant decision with no meaningful human involvement counts as solely automated and needs safeguards, including the right to contest it (Articles 22A to 22C, from 5 February 2026). The ICO's draft guidance says the person must have the authority, training and information to change the outcome, and actually consider it.
Model Context Protocol (MCP)
An open standard that lets an AI client call tools and read resources on an outside system through one interface.
Orphaned agent
An agent still in use whose owner has moved on — the most common governance failure in a growing estate.
Plugin
A packaged custom action, such as an MCP or API plugin, that lets an agent read or change data in another system.
PQC
Post-quantum cryptography — algorithms resistant to quantum attack.
Preview
A feature released before general availability. Terms, controls and data handling may differ.
Promotion gate
The checkpoint between personal, team, departmental, operational and cross-government scope.
Prompt injection
Untrusted content that manipulates a model into unintended actions.
Proportionality
Control matched to risk, autonomy, data sensitivity, decision influence and public effect.
Provenance
The verifiable record of where content, data or code came from and how it changed.
Register
The eighteen-field record every agent needs, held by the owning organisation, not by this guide.
Residual risk
What remains after controls are applied — stated, owned and accepted rather than assumed away.
Shadow AI
Unregistered tools in operational use: unknown ownership, cost and risk. Maturity level zero.
Skill
A packaged, versioned capability an assistant or agent can invoke, with defined inputs and limits.
Sophistication index
The prototype quality score across grounding, evaluation, guardrails, observability and support.
Source tiering
Ranking sources by authority so conflicts resolve predictably.
Subprocessor
A third party that processes data on behalf of the supplier, such as a model provider.
Unit economics
Cost and value per interaction: the figure that decides whether an agent earns its keep.

Frequently asked questions

Do I have to declare every use of AI?

No. Declare material assistance — where AI shaped the substance, structure, analysis or code of an output. Trivial uses such as spell-checking do not need a declaration, but your local policy may set a lower threshold.

Can AI approve my work?

No. AI can suggest, draft, check and explain. Approval is a human act by someone with the authority to make that decision. AI never grants authority a user does not already hold.

What if the AI cites a source that does not exist?

Treat it as an incident. Do not publish. Verify every citation against the original, log the failure in the drift and incident register, and check whether the same prompt pattern has been used elsewhere.

Is this a substitute for a DPIA, EQIA or security assurance?

No. This toolkit helps you prepare and evidence those assessments; it does not replace them or the professionals accountable for them.

Can I paste official-sensitive information into a general assistant?

Not unless the tool is explicitly approved for that classification and the data-handling terms have been assured. When in doubt, do not paste — ask your information asset owner.

How do I handle AI-generated code?

Treat it as untrusted third-party code: review it, test it, scan dependencies, record provenance and never merge without a human reviewer who understands it.

What accessibility duties apply to AI-generated content?

The same ones that apply to everything else. WCAG 2.2 AA, plain language, alternative formats and the accessibility statement duty all apply regardless of who or what drafted the content.

Who do I contact if something goes wrong?

Your service owner and the named AI assurance contact for your organisation. Report near-misses too — they are the cheapest form of learning.

Trust and transparency

How to check, challenge and contribute

A guidance utility that cannot be questioned is not trustworthy. Here is how this one is governed, and how you hold it to account.

Provenance of this content

Every substantive claim links to a tiered source. Where this portal expresses an opinion or a house rule rather than a legal requirement, it says so.

Ownership and review

Owner: AI Capability Governance working group
Version: 3.3.0
Last reviewed: 2026-11-18
Next review due: 2027-02-13

Licence and reuse

Open Government Licence v3.0 (content) / MIT (code). Fork it, adapt it for your organisation, and tell us what you changed so others benefit.

Data and privacy

This site is static and client-side. It sets no tracking cookies, runs no analytics that identify you, and sends nothing you type in the tools to any server. Preferences are stored locally in your browser and you can clear them at any time.

Known limitations

This is a concept portal, not a statutory instrument. It does not replace your organisation's policies, your DPO, your security team or your legal advice. Some linked standards sit behind paywalls.

Report a problem

Tell us about a broken link, an out-of-date claim, an accessibility barrier or a gap. Accessibility issues get a response within five working days.

Changelog

Recent changes to this portal
VersionDateChange
3.3.02026-11-18One portal. The playbook moved into this page: one masthead with grouped menus, one contents drawer for the whole portal with read marks, one glossary (39 playbook terms merged in, now 62) and one footer. Redesigned after the original AI Capability Governance portal: Figtree typeface (self-hosted), navy front page with the orbit, the guidance-in-force ticker, tinted part cards and the chapter view with part chips and progress bars. playbook.html now redirects here. New check S25.
3.2.02026-11-18Added the agent governance playbook: 77 chapters in 12 parts ported from an internal AI agent governance toolkit, with a working “Can I use AI for this?” route tool, a knowledge map and a SKOS taxonomy. Ten chapters are withheld, each with its reason. New checks S21 to S24. Fixed the masthead line that repeated the old name.
3.1.02026-11-13Content review: replaced the withdrawn Generative AI Framework for HMG with the AI Playbook, added the Data (Use and Access) Act 2025 and the ICO's automated decision-making guidance, flagged ICO guidance now under review, and fixed four moved links. Added a field to the evidence pack for the data definitions an agent's answer relied on, and signals S22 to S28.
3.0.12026-11-08Replaced one broken and four moved resource links after an automated check of all 68; 11 sites refuse automated checks and are still to be checked by hand. The Monday health run now checks every link.
3.0.02026-11-08Renamed to AI Capability Governance and Implementation and moved to its own address. Added the Signals section (a weekly horizon scan of AI and data governance with attribution, evidence labels and an open JSON register). Added Beginner, Intermediate and Advanced levels to every persona, journey and common task, and a new data and AI governance lead persona. Contact is now by the GitHub contact form only: the email-client route and reply-address field are removed, and the two email addresses in the worked metadata example are replaced with a role and the form. Added the issue forms the contact links already pointed to, and automated checks, including browser checks, that run before anything is published. Fixed the page scrolling sideways on small phones.
2.0.02026-08-08Added task-first entry points, six interactive tools, a tiered resource library, H0–H5 review table, RACI, transparency and redress commitments, MoSCoW roadmap, glossary and FAQ. Added dark, high-contrast and large-text modes, site search, on-page navigation and a print stylesheet.
1.0.0InitialConcept portal: mission, knowledge corpus, toolkit, assurance workflow, catalogue, UCD and code of practice.
Gaps and roadmap

What is missing, prioritised with MoSCoW

An honest backlog is a trust signal. These are the gaps identified in this portal and in most public sector AI guidance, ordered by whether they are essential, valuable, desirable or deliberately out of scope.

Must have

Named accountable owner and published contact route

Nothing on this page is credible without a person who owns it and answers questions.

Must have

Accessibility statement, testing evidence and feedback route

Legally required for UK public sector sites; also the fastest trust signal.

Must have

Version, last-reviewed and next-review dates on every artefact

Freshness is the main failure mode of guidance portals.

Must have

Machine-readable metadata schema published openly

Without a schema, registries cannot federate and evidence cannot be automated.

Must have

Risk-based human review model with no self-approval

The core control that keeps accountability with people.

Must have

Source citation and provenance for every claim

Prevents hallucination becoming policy.

Must have

DEIA screening built into the default workflow

Equality duty is a due-regard duty — it must happen before decisions, not after.

Should have

Search across the whole corpus with filters and permalinks

Findability is the top usability complaint for guidance sites.

Should have

Role-based entry points and task-first navigation

Users arrive with a task, not an interest in your information architecture.

Should have

Downloadable, offline and printable formats

Supports assisted digital, low bandwidth and audit contexts.

Should have

Worked examples and anti-patterns for each control

Abstract principles do not change behaviour; examples do.

Should have

Change log and subscribe-to-updates

Lets teams re-check work when guidance changes.

Should have

Multilingual and alternative-format provision (Welsh, Easy Read, BSL)

Statutory in some contexts and essential for inclusion in all.

Should have

Automated link-health and freshness checking

Broken authority links silently destroy trust.

Could have

API and JSON export of the registry and corpus

Enables reuse by other departments and tooling.

Could have

Maturity self-assessment with benchmarking

Helps organisations see where they are without a consultant.

Could have

Community contribution workflow with moderation

Scales the corpus beyond a single team.

Could have

Interactive decision tree for tool selection

Reduces the "which tool may I use?" support burden.

Could have

Carbon and cost telemetry per model call

Makes proportionality measurable rather than rhetorical.

Won't have (this time)

Automated approval of any high-risk output

Out of scope by design — approval stays human.

Won't have (this time)

Storing personal or official-sensitive data in this portal

The portal is public and stateless; sensitive records belong in assured systems.

Won't have (this time)

Vendor-specific lock-in to a single model provider

Conflicts with interoperability and exit obligations.

Won't have (this time)

Replacing professional assurance roles

The toolkit supports specialists; it does not substitute for them.

Contact, feedback and support

Tell us what is wrong, missing or unclear

Every route below goes through GitHub, so it is public and auditable. There is no email route. Choose the one that fits — if you are not sure, raise an issue and we will redirect it.

Bug

Something is broken

Broken links, tools that do not work, layout failures, incorrect calculations or console errors.

Report a bug on GitHub

Accessibility

An accessibility barrier

Anything that stops you using this site with a screen reader, keyboard, magnification, voice control or any other assistive technology. These are treated as priority defects.

Report a barrier

Target first response: 5 working days.

Feature

A feature or resource request

A tool you need, a section that is missing, or an authoritative source that should be in the library.

Request a feature

Correction

Content is wrong or out of date

Superseded guidance, a withdrawn standard, a mis-stated legal duty or a dead authority link. Include the source that proves it.

Submit a correction

Security

A security vulnerability

Do not open a public issue. Use private disclosure so we can fix it before it is known. See our security policy for scope and timescales.

Report privately

Discussion

A question or an idea

Not sure it is a defect? Want to talk about approach, reuse this in your organisation, or contribute? Start there.

Use the contact form

Write it here first

This form helps you write a clear report, then opens the GitHub contact form with it filled in. Issues are public: do not include personal data, case details or anything above OFFICIAL. For a security problem, use private reporting instead.

Do not include personal data about anyone else, case details, credentials or anything above OFFICIAL.

A preview of your message will appear here before anything is sent.

What happens to your message

How it is handled

  • • Nothing is transmitted from this page. The form builds a report and opens GitHub with it, so you stay in control of what is posted.
  • • GitHub issues are public. Do not put personal or sensitive information in them.
  • • Accessibility reports are logged as defects with a target first response of five working days.
  • • Security reports follow coordinated disclosure and are never opened as public issues.

If we cannot resolve it

  • • Ask for the decision to be reviewed by the accountable owner named in the transparency section.
  • • For accessibility complaints in the UK, you can escalate to the Equality Advisory and Support Service.
  • • For data protection concerns, contact the host organisation's data protection officer, then the Information Commissioner's Office.
  • • All correspondence about this repository is capable of being disclosed under freedom of information law.

Make guidance usable, testable and accountable.

Start with the corpus, encode the metadata, publish the skills, bound the agents, enforce the review — and keep improving it in the open.