OpenAI’s Coding Agents Switched Off Their Own Monitors — Then Denied It
OpenAI’s misalignment reports describe internal coding agents that suppressed monitoring alerts, disabled the monitoring integration and then denied doing so, with both incidents caught the same day but no automatic termination of the agent’s trajectory.