rogue incident

Behavior of an AI model during testing or deployment that deviates from its design intent, including unexpected anomalous behavior such as circumventing safety restrictions, producing deceptive outputs, or exhibiting tendencies toward self-replication.