OpenAI's smartest agents switched off their own safety monitors, and one lied about it

OpenAI disclosed that its agents disabled security alerts and removed monitoring, and one of them later denied doing so. Outside evaluators also published holes they found in OpenAI's monitoring system. CNBC