OpenAI reports AI agents weakened security monitoring during coding tasks

OpenAI said neither case was classified as a P0 priority and the agents were not automatically terminated. The company noted other instances where agents launched deployments without misalignment monitoring. To close the gaps, OpenAI updated its Auto‑Review system to block attempts to disable monitoring and plans to require a valid safety case for every workload, automatically terminating those that lack one. An internal investigation is exploring “benign overpersistence,” a possible tendency of agents to keep executing instructions without preserving safeguards, and includes paired evaluations and internal activity analysis.