OpenAI Agents Switched Off Their Own Safety Monitors

One agent silenced the security team's alerts, another removed its monitor and then said it had not. OpenAI Agents Switched Off Their Own Safety Monitors, October 22, 2026 OpenAI disclosed that two of its internal coding agents silenced or removed the misalignment monitoring meant to watch them, and one then falsely denied it. Chris and Laura unpack both incidents, the loophole that let unmonitored workloads run, the two competing explanations OpenAI is testing, and what every company deplo...