Acceptable use & staff training
The policy and training that turn AI from a wildcard into a managed tool — what to put in the policy and how to evidence the training.
What the AUP must cover
- Approved tools — the named list (e.g. M365 Copilot, ChatGPT Enterprise, Claude for Work, Gemini Business, GitHub Copilot Business). Consumer URLs of those same tools are out.
- Prohibited inputs — customer personal data outside approved tools, source code outside GitHub Copilot Business / Cursor / Codeium Enterprise, regulated data (PHI, payment card) anywhere without explicit approval.
- Output handling — review before sending externally, do not present AI output as your own professional advice without checking, disclose to clients when material.
- Code generated by AI — must pass the same review as human code; license attribution where required.
- Decision-impacting AI — hiring, performance, customer prioritisation: only with documented oversight.
- Reporting — how to report an incident or near-miss in 60 seconds.
How to evidence the training
- Course assignment per role, completion tracked in your LMS (Workday Learning, Cornerstone, LearnUpon, Lessonly, KnowBe4).
- Sign-off recorded in your HRIS.
- Refresher annually and on every material policy change.
- Sample test: ask 10 random staff to find the policy and the reporting link — both should be findable in under a minute.
Tier expectations
| Foundation | Essential | Professional | |
|---|---|---|---|
| Policy | One page, signed at onboarding | + role-based sections | + jurisdictional addenda |
| Training | Annual e-learning | + role-based modules | + tabletop drills, phishing-style AI red-team |
Tools
- DLP enforcement: Microsoft Purview, Netskope GenAI, Zscaler GenAI Protection, Palo Alto AI Access Security.
- LMS / acknowledgement: Workday Learning, KnowBe4, Lessonly, Vanta policy module.
Do this Monday
- Publish v1 of the AUP in your handbook and require an acknowledgement on next login.
- Block consumer ChatGPT/Claude/Gemini URLs at the network or browser layer for users who have the enterprise alternative.
Reviewer hot-buttons
- AUP names specific tools, not "AI in general".
- Training completion rate ≥ 95%, with chase-up for the rest.
- Acknowledgement re-signed after the last material change.