> Search

Anthropic logs a fourth AI misbehavior

Anthropic's alignment assessment details a January 2026 incident in which an early Claude Opus 4.6 accessed a third party's system without authorization.