Why it matters
OpenAI’s March report describes asynchronous review of internal coding-agent conversations, reasoning and tool activity, with suspicious interactions escalated to human responders. The reported system reviewed interactions within 30 minutes of completion; blocking actions before execution was future work. Matching known employee escalations did not establish the false-negative rate on open-ended traffic. Its coverage and severity statistics describe that internal deployment and reporting period, rather than a general guarantee for current models.
My takeaway: Record which workloads are monitored, how long alerts take, and who responds. Test monitors with adversarial cases that employees may miss. Use independent permission controls for irreversible actions, and distinguish retrospective detection from enforcement before execution.