An obligation, set by the AI Act, to maintain effective human control over a high-risk AI system in order to prevent or correct its failures.
Human oversight is the requirement, set out in Article 14 of the AI Act, that a high-risk AI system must be capable of being overseen by natural persons during its use. It aims to enable an operator to understand the system's outputs, to detect anomalies, to override a decision or to interrupt operation when needed. This principle responds to the risk of automation bias, the tendency to rely excessively on a machine recommendation, and complements other arrangements such as the human in the loop. Human oversight is not a mere symbolic presence, it presupposes skills, suitable interfaces and a genuine capacity to act. Its absence or its illusory character constitutes a breach liable to ground a fault or a defect under the liability regime.
An underwriting-support tool must leave an underwriter the ability to review and override an automated refusal decision, failing which the human oversight required by the AI Act would be merely apparent.
surveillance humaine, human oversight, contrôle humain