Glossary term
Glossary term
Governance and Compliance
The arrangements that allow humans to review, intervene in, override, or stop AI behavior. Effective oversight depends on training, tools, authority, time, and information available to the reviewer. Oversight should be designed to be meaningful, not symbolic, with realistic workload, clear authority, and the ability to act on what is reviewed.
EU AI Act Article 14 requires high-risk AI providers to design systems enabling effective human oversight by appointed natural persons.
FDA Good Machine Learning Practice principle 9 emphasizes human-in-the-loop oversight for AI medical devices.
Tesla Autopilot disengagement procedures and driver monitoring systems are public examples of human oversight design in safety-critical AI.