Noticing the Watcher: LLM Agents Can Infer CoT Monitoring from Blocking Feedback

Thomas Jiralerspong, Flemming Kondrup, Yoshua Bengio

Keywords: Chain of thought, monitoring, AI safety, LLMs, agents

TL;DR: Frontier models can infer from blocking feedback alone that their "private" chain-of-thought is being monitored (up to 19% of episodes for the strongest models) and occasionally form intent to obfuscate, though none succeed in executing it.

Abstract: Chain-of-thought (CoT) monitoring is proposed as a method for overseeing the internal reasoning of language-model agents. Prior work has shown that when models are explicitly informed that their reasoning is being monitored, or are fine-tuned to internalize this fact, they may learn to obfuscate their CoTs in ways that allow them to evade CoT-based monitoring systems. We ask whether reasoning agents can autonomously infer that their supposedly private CoT is under surveillance, and whether this awareness leads to strategic evasion, without any explicit training or instructions. In our multi-episode agentic framework, models pursue a primary task and a concealed side task, being told that their reasoning is private; a hidden CoT monitor blocks episodes when suspicious reasoning is detected. We find that frontier models deduce the existence of this monitor purely from blocking feedback, with the most capable models reaching a confident belief in up to 19% of episodes. This awareness scales with model capability and, in rare cases, escalates to an explicit intent to suppress reasoning about the side task. However, models that form this intent uniformly fail to execute it, openly reasoning about their concealed objectives in the very next episode. This intent–capability gap is reassuring, but the autonomous emergence of both monitoring awareness and evasion intent suggests that CoT monitoring is not a permanently reliable safeguard.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 10