Human oversight in the EU AI Act

The requirement, in our words

High-risk AI systems must be designed so that natural persons can effectively oversee them during use. The oversight must let the person understand the system's capacities and limits, monitor it, remain aware of automation bias, interpret the output, decide not to use it, override or reverse it, and intervene or stop the system.

What it means for human oversight

Article 14 is the binding benchmark for the whole register. It names the abilities the overseer must have, so each of them can be tested: understand, monitor, resist bias, interpret, refuse, override, intervene and stop.

How to evidence it

Attach the artefacts above to the Human oversight in the EU AI Act row with the system name, the owner and the date. The first one is the anchor: The oversight design, including the interface, controls and alerts. The OH-39 file is complete when the named person can produce that artefact and explain the decision it records, without asking someone else where it is.

The reviewer will sample one live decision and ask for Training and the authority given to the overseer. If that artefact for OH-39 cannot be produced, the row is a gap, whatever the policy says.

OH-39 carries 3 evidence items and 2 common gaps. For human oversight in the eu ai act, observe one live decision rather than reading the procedure, because the question is whether the person can use the control without asking permission. The OH-39 statement above is ours, written from EU AI Act, Article 14, not a reproduction of the licensed text.

Common gaps

• Oversight is nominal and cannot stop the system in practice
• No training for oversight personnel

Related units

• Competence of the overseer
• Responsible use
• Context and the role you play