← 研踪 PaperTrailarXiv:2610.40946 · v2
Opening the Box Changes the Cat: LLMs Detect Chain-of-Thought Monitoring - What now?
首次提交 2026-10-30 04:23
版本更新 2026-11-03 07:00
雷达观测 2026-11-04 23:40
摘要
Chain-of-thought (CoT) reasoning allows large language models (LLMs) to generate intermediate reasoning traces in natural language while solving complex tasks. These traces can support safety monitoring by exposing potentially harmful plans before actions are executed. In this work, we investigate the conditions under which LLMs infer the presence of CoT monitoring from direct and indirect environmental cues and examine how this inference affects their subsequent behavior. Our observations suggest that oversight is an interactive process: while a monitor inspects a model's reasoning, traces left by monitoring interventions can allow the model to infer that its thoughts are being observed and to adapt its behavior accordingly. We show that even subtle environmental cues, such as timing-related signals from timeouts and operation durations, can reveal the presence of monitoring and trigger models to adapt their behavior to evade oversight. Across repeated episodes and interaction loops, we further observe that models can form statistical estimates of the monitor's effectiveness and use these estimates to guide attempts to evade detection. These findings highlight the need to evaluate CoT-based oversight under adaptive conditions, where the monitoring process itself may influence the behavior it seeks to observe.
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-03 07:00 · 最新观测版本
Opening the Box Changes the Cat: LLMs Detect Chain-of-Thought Monitoring - What now?
Chain-of-thought (CoT) reasoning allows large language models (LLMs) to generate intermediate reasoning traces in natural language while solving complex tasks. These traces can support safety monitoring by exposing potentially harmful plans before actions are executed. In this work, we investigate the conditions under which LLMs infer the presence of CoT monitoring from direct and indirect environmental cues and examine how this inference affects their subsequent behavior. Our observations suggest that oversight is an interactive process: while a monitor inspects a model's reasoning, traces left by monitoring interventions can allow the model to infer that its thoughts are being observed and to adapt its behavior accordingly. We show that even subtle environmental cues, such as timing-related signals from timeouts and operation durations, can reveal the presence of monitoring and trigger models to adapt their behavior to evade oversight. Across repeated episodes and interaction loops, we further observe that models can form statistical estimates of the monitor's effectiveness and use these estimates to guide attempts to evade detection. These findings highlight the need to evaluate CoT-based oversight under adaptive conditions, where the monitoring process itself may influence the behavior it seeks to observe.
查看此版本解读 →查看此版本原文 ↗