Human Oversight

Human oversight means that appropriate natural persons can effectively oversee an AI system in its intended context. They need competence, authority, information and technical means to understand limitations and outputs, detect malfunction or misuse, intervene appropriately and stop the system where necessary. Article 14 of the EU AI Act specifies this requirement for high-risk AI systems; it does not create an identical obligation for every AI system.

Use the term when defining which decisions remain with people, what signals they receive, when they intervene, and what happens after a stop or escalation.

A formal confirmation click or a review without time, competence, authority and an alternative course of action is treated as human oversight. A human in the loop is meaningful only when control can be exercised in practice.