When an AI agent states something confidently that no real source supports.
The answer is fluent and specific and sounds exactly as certain as a correct one. There is nothing true behind it: a policy that does not exist, an approval no system issued, a number that is wrong. The dangerous ones look identical to correct answers, so nobody flags them until the damage is done.
Related: AI Agent Drift, AI Agent Monitoring, AI Agent Audit Trail
See also: What Is AI Agent Hallucination