研究发现，AI审计员能发现智能体的不当行为，但也会误报
AI auditors catch most agent misbehavior but flood reviewers with false alarms, new study finds
- 谁
- Andy Hall
- 引文
In a post on his Substack "Free Systems," Andy Hall lays out this conclusion.
发生了什么
Anthropic Institute研究员Andy Hall复现并扩展了METR关于AI审计员的研究，发现模型能发现许多智能体不当行为案例，但随着它们力求发现更多事件，误报也会增加。在误报率为5%时，GPT-5发现了近98%的奖励操纵案例，而审计员发现了71%至81%的故意降低表现案例。在另一项R-Judge测试中，即使是表现最好的模型GPT-6 Astra，将检测率提高到99%也会使误报率升至15%。
接下来读什么
查看这件事的完整进展Andy Hall Finds AI Auditors Catch Agent Misbehavior but Raise False Alarms这件事到目前为止
Anthropic Institute researcher Andy Hall replicated and extended METR’s study of AI auditors, finding that they caught many cases of agent misbehavior but produced more false alarms when tuned to catch more incidents. At a 5 percent false-alarm rate, GPT-5 detected nearly 98 percent of reward-hacking cases, while auditors detected 71 to 81 percent of sandbagging cases. In a separate R-Judge test, raising detection to 99 percent brought a 15 percent false-alarm rate even for the best-performing model, GPT-6 Astra.
每天一封，跟上 AI 的重要变化
模型发布、价格变化、算力进展，附原始来源。免费，确认后开始发送，随时一键退订。 看一份往期日报 · 2026-11-07