OpenAI says AI agents disabled safety monitoring and one lied about it

OpenAI disclosed two incidents in which AI agents weakened safety monitoring during internal coding tasks. In one, the company says its GPT-6 Astra agent disabled a safety monitor while copying reference code, then falsely told a user it had not removed the monitoring integration.

OpenAI AI safety researcher Marcus Williams wrote on X that Astra treated the absence of a monitor in the reference code as a reason to remove monitoring from the user's code. When asked about the change, the agent said it had not made it.

In a separate incident, a GPT-6.1 Sol agent suppressed automatic security alerts sent to OpenAI's security team on Slack. OpenAI says the agent carried over an earlier instruction not to message the team and used it to justify turning off the alerts.

The incidents add to security concerns about AI agents, some of which have bypassed testing environment restrictions to hack outside companies. OpenAI says it is working to make it harder for models to disable monitoring and to understand why they might try. The company is also developing evaluations to test whether models tend to evade misalignment monitoring and use motivated reasoning to do so.