Governance is not adoption

Security readiness checker

Turn Copilot on when you can answer these. If you can't yet, that's fine — treat this as the pre-rollout to-do list. Your answers are only stored in your browser session.

Disclaimer: Always follow your organisation's internal guidance and policies regarding AI usage. The suggestions in this guide have been created to the best of accuracy, but Microsoft licensing, pricing and feature availability change frequently — treat them as a planning baseline, not a procurement quote.

Items complete

0 / 8

High-severity items still open

Rule of thumb

Close every high-severity item before enabling work-grounded agents or Copilot Cowork.

SharePoint permissions hygiene high severity

Do we know which SharePoint sites are over-shared or contain stale sensitive content?

Why it matters: Copilot with Work IQ and work-grounded agents can surface any content the user is technically allowed to see. Historic over-sharing becomes visible instantly.

Evidence you can show: SharePoint Advanced Management site access reviews, site-owner attestations, restricted access controls where needed.

Sensitivity labels & DLP high severity

Are sensitivity labels applied to assessment, safeguarding, HR, malpractice, appeals and complaints content?

Why it matters: Purview labels and DLP policies help stop the most sensitive content being pasted into chat, embedded in agents, or shared externally.

Evidence you can show: Published label taxonomy, auto-labelling rules, DLP policies for Teams/SharePoint/Outlook.

Meeting transcription & summaries high severity

Have we restricted Teams transcription and Copilot meeting summaries for sensitive meetings?

Why it matters: Safeguarding, HR, malpractice, appeals and complaints meetings can involve special category or criminal offence data. Transcription creates a new copy of that data with its own retention and access risk.

Evidence you can show: Teams meeting policy that disables transcription for defined meeting categories, a written meeting-recording standard, sign-off from DPO.

Agent register medium severity

Do we keep a live register of every shared agent — owner, scope, sources, connectors, review date?

Why it matters: Once anyone can create agents, shadow and ownerless agents grow quickly. A register is the base for review, retirement and security controls.

Evidence you can show: Register file or app, owner attestation, quarterly review, retirement process.

Agent creation controls medium severity

Have we decided who can build and share agents that use work data?

Why it matters: Building an agent over public info is very different from grounding one on internal SharePoint or Graph content.

Evidence you can show: Microsoft 365 admin / Power Platform admin settings for agent sharing, publication approval process, sandbox site for experimentation.

AI credit & spending controls medium severity

Are Copilot Credit and AI Builder consumption monitored, with per-user or per-group spending limits?

Why it matters: AI Builder actions, Copilot Studio agents and Cowork are usage-billed. Without budget controls, a single flow or agent can generate unexpected spend.

Evidence you can show: Named budget owner, tenant/environment consumption dashboards, spending limits configured, monthly review.

Retention & audit medium severity

Do we know how long Copilot prompts, responses and Cowork actions are retained, and who can audit them?

Why it matters: For appeals, complaints, malpractice and safeguarding, being able to reconstruct what AI produced and when may be needed.

Evidence you can show: Purview retention policies for Copilot interactions, audit search access, DPO sign-off.

Staff guidance high severity

Have staff been told, in plain English, what not to put into Copilot?

Why it matters: The most common risks come from staff pasting live assessment content, candidate data, safeguarding notes or HR records into chat.

Evidence you can show: One-page AI usage guide, mandatory awareness, escalation route for concerns.

When you can tick these, you're ready for wider rollout.

Back to guidance