Glossary
EU AI Act Article 14 (Human Oversight)
Definition: Article 14 of the EU AI Act requires that high-risk AI systems be designed so a human can effectively oversee them, including the ability to intervene in or stop the system, a requirement most current autonomous-agent deployments are not built to meet.
“Human oversight” sounds like a dashboard. Article 14 means something narrower and harder: a tested capability to actually terminate a misbehaving agent, mid-action, with its state captured for review, not just a monitoring alert someone can ignore. The gap this closes is real and measured: one industry survey found 60% of organisations deploying AI agents have no way to terminate one that starts misbehaving. Monitoring is not containment. An agent that runs 847 transactions before anyone intervenes was being watched the entire time, watching was not the control that was missing.