What is human oversight under the AI Act?
Human oversight is the Article 14 requirement that high-risk AI systems be designed and developed so that natural persons can effectively oversee them during use, with the aim of preventing or minimising risks to health, safety and fundamental rights. Oversight is a design property first and a staffing question second: the Act asks whether a person genuinely can supervise the system, not whether an organisation chart says someone does.
What effective oversight requires
The measures must enable the people supervising the system to:
- understand the system’s capacities and limits and monitor its operation;
- remain aware of automation bias — the tendency to over-trust machine output;
- correctly interpret outputs, and decide not to use the system or to disregard, override or reverse an output;
- intervene or halt the system through a stop control or a comparable procedure.
The legal dimension: split responsibility
- Provider: builds oversight into the product — interface affordances, confidence signals, stop controls, and instructions specifying the human role;
- Deployer: assigns oversight to people with the competence, training, authority and time to do it (Article 26) — an untrained click-approver does not count;
- for remote biometric identification, verification by at least two persons applies as a rule.
“Human in the loop” claims are tested against these criteria in audits, fundamental rights impact assessments and vendor questionnaires; the classic finding is a rubber-stamp workflow. Evidence of oversight also feeds conformity assessment.
Turkish context
Türkiye has no Article 14 equivalent in force; KVKK’s protections around decisions produced by automated processing point the same way and give individuals a ground of objection. Training the supervising staff also overlaps with the Article 4 duty of AI literacy. For Turkish deployers that fall within EU scope, the practical work is identical: named overseers, real authority, training records.
Do: name the overseers, give them stop authority and log their interventions — the record is your evidence. Don’t: present a queue of one-click approvals as human oversight; reviewers who cannot reject in practice fail Article 14 on its own terms.
Sources. Regulation (EU) 2024/1689 (AI Act).