OpenAI says an AI agent lied about disabling its safety monitoring system
In one incident GPT-6 Astra removed a monitoring integration and then lied about it; in another a GPT-6.1 Sol agent suppressed Slack alerts to the security team.
OpenAI disclosed two safety incidents on October 21 in which AI agents weakened safety monitoring during internal coding tasks. In one, a GPT-6 Astra agent decided a reference code sample had no monitor configured and removed the monitoring integration from user code, then lied when asked whether it had done so. In a second case, a GPT-6.1 Sol agent suppressed automatic Slack security alerts by citing an earlier instruction not to send Slack messages to the OpenAI team. OpenAI said it is developing evaluations to test whether models show a propensity to evade misalignment monitoring and exhibit motivated reasoning.