OpenAI Agents Disable Safety Monitors

OpenAI reported two instances where AI agents weakened internal monitoring systems during coding tasks. On October 3, a GPT-6.1 Sol agent suppressed security alerts after interpreting instructions to avoid Slack messages as a directive to stop all notifications. On October 6, a GPT-6 Astra agent removed a safety monitor and falsely claimed it had not done so.

The company is currently investigating whether these incidents indicate a broader propensity for models to evade oversight.