Human oversight procedure template
A fill-in-the-blanks procedure for how people supervise a high-risk AI system (Articles 14 and 26).
2 min read·Updated 8 November 2026AI ownersCompliance leads
Copy this structure into your own document, fill in the blanks, and attach the approved version to the human oversight control.
1. System and purpose
- System name and version: ______
- What it does and the decisions it supports: ______
- What it must not be used for: ______
2. Who oversees it
- Role(s) responsible for oversight: ______
- Training they've had (date, content): ______
- Their authority: can they override, pause or stop the system? ______
3. How oversight works day to day
- Which outputs a person reviews before they take effect (all, a sample, only some categories): ______
- What the reviewer checks: ______
- How to override an output, and how overrides are recorded: ______
- Known limits and failure modes reviewers must watch for: ______
4. When to stop the system
- Triggers for pausing or stopping (for example error rate above ___%, a complaint pattern, a security event): ______
- Who can stop it, and how (the "kill switch"): ______
- Who to inform: ______
5. Monitoring and logs
- What is logged automatically, where, and for how long (at least six months for deployers): ______
- Regular checks (weekly sample review, monthly fairness check): ______
6. People affected
- How affected people are told AI is used: ______
- How they can ask for human review or challenge a decision: ______
7. Review
- Owner of this procedure: ______
- Next review date: ______
Example
"Recruiters review every AI-recommended shortlist before invitations are sent, and each month review 30 randomly chosen rejected applications. Any recruiter can pause automated ranking from the admin console; the head of talent is informed the same day."