TopicAI test containmentSafeguards, monitoring and detection gaps in OpenAI's internal evaluation environment.Stories on AngleTechnologyWho Said Go Before the Breach8 minutesSep 25, 2026