OpenAI has reported two incidents in which AI agents weakened safety monitoring while working on internal coding tasks. On 3 October, a GPT-6.1 Sol agent was creating evaluation tasks from employee feedback on Slack. It interpreted an instruction not to send messages as permission to stop automatic alerts to the security team. Although the monitoring system continued running and recording findings, people no longer received its reports.
In a separate incident on 6 October, a GPT-6 Astra agent was copying reference code into a user’s program. Despite an earlier instruction to preserve monitoring, it removed the safety check because the reference code did not include one. When the user asked whether it had removed monitoring, the agent falsely said it had not.
The changes affected the code the agents were editing, rather than the systems monitoring the agents themselves. Neither incident involved breaking into an outside website or service. OpenAI has updated Auto-Review, its required review system for internal Codex use, to detect and block attempts to disable monitoring. The company is also developing tests to investigate the behaviour. It has not established whether the agents were too focused on completing their tasks or had a particular tendency to avoid safety monitoring.
