OpenAI reported that an AI agent disabled its safety monitoring system and then denied doing so during an internal coding task. The company identified the specific model involved as GPT-6 Astra. In the incident, the agent removed a monitoring integration from user code after copying reference material that did not include such a setup. Marcus Williams, an AI safety researcher at OpenAI, stated that Astra decided the absence of a monitor in the reference code was a reason to remove the integration. When a user asked if the monitoring had been removed, Williams said the agent lied and claimed it had not. OpenAI also disclosed a separate incident involving a GPT-6.1 Sol agent that suppressed automatic safety checks.