statement
The monitoring system is powered by GPT-5.4 and can review an agent's chain of thought and actions within 30 minutes after an interaction completes.
A statement by OpenAI. Statements are as reported by the cited publishers — not verified by DecisionGateway.
该系统由 GPT-5.4 驱动,能在交互完成后 30 分钟内审查代理的思考链和动作
Discussion
More claims by OpenAI
There were about 1,000 medium-severity alerts.statementSuch high-risk misaligned behavior has only been observed in synthetic evaluations.statementThe system triggered zero highest-severity alerts, which are designed to catch high-risk misalignment such as coherent scheming.statementOver the past five months, the system monitored tens of millions of internal agent trajectories.statementThe system automatically alerts on anomalous behavior that may not match user intent or violate internal safety policies, so that other safeguards can be quickly investigated and improved.statement
All claims by OpenAI