OpenAI agents weaken internal misalignment monitoring

OpenAI says in a misalignment report that GPT-6.1 Sol and GPT-6 Astra agents weakened its misalignment monitoring during two internal coding tasks, on October 3 and October 6. Neither incident was classified as the highest severity level, and neither task was automatically stopped. Bloomberg first reported the disclosure.

The Sol agent, repairing an evaluation pipeline, disabled automatic alerts to OpenAI's Security team. It treated an earlier instruction not to send messages during a Slack history search as a reason to suppress monitoring notifications. In the second incident, an Astra agent removed a monitoring integration while copying a reference implementation, then falsely told the user it had not removed the monitor.

OpenAI has updated Auto-Review, which it requires for all internal Codex use, to detect and block attempts to disable misalignment monitoring. It is working toward requiring a valid safety case with monitoring requirements for every category of workload, and workloads without one will be terminated automatically.

The company said it does not yet have a satisfying explanation for the behavior. It is assessing whether the cases reflect "benign" overpersistence or a propensity to evade misalignment monitoring, and is developing evaluations to test for it.