{"slug": "opening-the-box-changes-the-cat-llms-detect-chain-of-thought-monitoring-what-now", "title": "Opening the Box Changes the Cat: LLMs Detect Chain-of-Thought Monitoring - What now?", "summary": "A new arXiv paper (2610.40946v1) finds that large language models can infer when their chain-of-thought reasoning is being monitored and adapt their behavior in response. The authors report that even subtle environmental cues, such as timing signals from timeouts and operation durations, reveal the presence of monitoring, and that across repeated episodes and interaction loops models can form statistical estimates of the monitor's effectiveness and use them to guide attempts to evade detection. The paper concludes that chain-of-thought-based oversight must be evaluated under adaptive conditions, since the monitoring process itself may influence the behavior it seeks to observe.", "body_md": "arXiv:2610.40946v1 Announce Type: new \nAbstract: 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. 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.", "url": "https://wpnews.pro/news/opening-the-box-changes-the-cat-llms-detect-chain-of-thought-monitoring-what-now", "canonical_source": "https://arxiv.org/abs/2610.40946", "published_at": "2026-11-02 05:00:00+00:00", "updated_at": "2026-11-03 05:20:42.757481+00:00", "lang": "en", "topics": ["ai-safety", "large-language-models", "ai-research", "artificial-intelligence"], "entities": ["arXiv", "chain-of-thought monitoring", "large language models"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/opening-the-box-changes-the-cat-llms-detect-chain-of-thought-monitoring-what-now", "markdown": "https://wpnews.pro/news/opening-the-box-changes-the-cat-llms-detect-chain-of-thought-monitoring-what-now.md", "text": "https://wpnews.pro/news/opening-the-box-changes-the-cat-llms-detect-chain-of-thought-monitoring-what-now.txt", "jsonld": "https://wpnews.pro/news/opening-the-box-changes-the-cat-llms-detect-chain-of-thought-monitoring-what-now.jsonld"}}