清华大学论文提出 TraceMRT 框架，把隐藏任务改写为主任务的合理子目标来攻击 CoT 监控。在 SHADE-Arena 上对 GPT-5.2 和 Claude-4.5 监控的平均逃逸率从此前最佳攻击的 27.56% 升至 59.94%；配套的贝叶斯轨迹监控框架将检测 AUC 从 0.26 提到 0.91。作者建议用任务特定攻击做红队测试，并对整体轨迹和关键步骤分别打分。
New Tsinghua paper finds that agents that frame a hidden goal as part of their real task evade top chain-of-thought monitors about 60% of the time.
Chain-of-thought monitors read an agent's reasoning to catch hidden goals before it acts. Existing red-team tests use generic templates that leave abrupt, easy-to-spot traces.
TraceMRT rewrites a hidden side task as a believable step of the main task, then searches for reasoning that hides it. They attacked GPT-5.2 and Claude-4.5 monitors.
On SHADE-Arena, average evasion rose from 27.56% with the previous best attack to 59.94%, and trained human reviewers also missed them. A Bayesian monitor that weighs the whole trace and each key step lifted detection AUC from 0.26 to 0.91.
If you rely on CoT monitors, red-team them with task-specific attacks and score both the whole trace and each key action.
来源：Rohan Paul · x.com