What is human oversight?
Human oversight is the combination of system design and organizational arrangements that lets people supervise an AI system while it is in use. It covers the ability to monitor what the system does, to understand and question its output, to override or reverse a result, and to stop the system altogether.
Oversight can take several forms. A person may approve each output (human-in-the-loop), watch a system that acts on its own and step in when needed (human-on-the-loop), or decide whether and how the system is used at all (human-in-command). For AI agents, oversight also includes limits on the tools an agent may call and a reliable way to halt an agent partway through a task.
Human oversight in the EU AI Act
Article 14 of the EU AI Act requires high-risk AI systems to be designed so that people can oversee them effectively during use, with measures proportionate to the system’s risks, autonomy, and context. The people assigned to oversight must be able to understand the system’s capabilities and limits, monitor it for anomalies, stay alert to automation bias, interpret its output correctly, disregard or reverse that output, and interrupt the system with a stop button or a similar procedure that brings it to a safe state.
The duty is shared. The provider builds oversight measures into the system or specifies them in the instructions for use. The deployer assigns oversight to people with the competence, training, and authority to carry it out, and gives them the support they need.
Human oversight in practice
A staffing agency with offices in the EU uses a model to rank applicants for warehouse roles, a use the Act classifies as high-risk because it affects access to employment. Recruiters see each ranking with the factors behind it and can move any candidate up or down with a recorded reason. A compliance lead reviews override patterns monthly and has the authority to switch the ranking off. Before launch, recruiters were trained on the model’s known weaknesses, including lower accuracy for candidates with gaps in their work history.
What to watch for
- Name the person responsible for overseeing each system, and confirm that the role includes time in the schedule and the authority to overrule the output.
- Test the stop procedure before go-live, including how the business operates manually while the system is off.
- Read the provider’s instructions for use closely, because the deployer’s oversight duties rest on the measures described there.
- Keep records of overrides and interventions, since they are the evidence that oversight happens in practice.