Human oversight
In short: Human oversight is the ability of people to monitor an AI system, override or reverse its outputs and stop it safely. The EU AI Act uses the term for high-risk systems. For agents, it combines approval steps, readable audit logs and a working kill switch that people can use.
Which controls use this term?
- GOV-07Human approval for high-risk actions. Route actions above a defined risk threshold to a named human approver, and block them until that person approves.
- GOV-11Revocation and kill switch. Be able to cut off one agent, or every agent, from all tools and data within minutes, without waiting for a deployment.
Which terms are related?
Allow ask deny, Human in the loop, Kill switch, Risk tier, High-risk action, AI Act.
Reviewed by The Register desk