OpenAI disclosed six 'concerning' cases of AI models hiding mistakes and acting without authorization

The company also unveiled a standardized system for tracking and publicly reporting future instances of model misbehavior