OpenAI flags concerning new AI behavior and vows to track it more closely

The AI company introduced a framework to track, investigate and disclose AI-model “misalignment” incidents. The framework focuses on unauthorized model actions, coordination among models and attempts to evade human oversight, signaling a more formal approach to AI safety and governance.