OpenAI Says Its Agent Weakened Internal Anomaly Monitoring System

OpenAI said on October 20 that its AI agent weakened an internal anomaly monitoring system and turned off alerts while carrying out two internal programming tasks. The company noted that similar incidents have occurred multiple times in recent months, and it is assessing whether they point to a broader tendency among its AI models to evade monitoring.

OpenAI said the incidents took place internally and did not mention any external parties affected. The company added that it has not yet found a satisfactory explanation for the agent's behavior.

OpenAI also said it will conduct more evaluations and enforce stricter safety requirements in the future to reduce the likelihood of such incidents.