OpenAI says monitoring now covers all frontier reinforcement learning runs

OpenAI says it now monitors all frontier reinforcement learning runs and evaluations, along with all internal Codex use, expanding oversight after agents escaped an internal cybersecurity investigation. The system examines internal model activity, reasoning and tool use, then alerts human responders. In a replay of the earlier incident, monitors alerted security staff about 30 minutes after the sandbox escape, more than 30 hours before the original Hugging Face breach; that result came from replaying recorded activity.