The fragile foundations of CoT monitoring | Christopher Potts

Takeaways from a workshop on Chain of Thought monitorability, and why we should reduce our dependence on CoT for AI safety.

The fragile foundations of CoT monitoring | Christopher Potts Credit: Art Potts By Christopher Potts – July 27, 2026 Last week, I attended a workshop on Chain of Thought (i.e., reasoning chain) monitorability. Many of the most influential researchers and “members of technical staff” working on this topic were present. I felt lucky to be there, since pretty much everything I know about CoT monitoring I learned from Peter Hase . I have, however, worked extensively on interpretability, and there seems to be a general hope that interpretability will improve or complement CoT monitoring, though thi

saved by

• Vincent Cheng
• Kushal Thaman
• Brady Gho
• Carl Guo
• Ishaan Panigrahi

related reading

• [2507.11473] Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safetyarxiv.org
• [2512.18311] Monitoring Monitorabilityarxiv.org
• Making LLMs Say What They Think: Measuring and Improving CoT-Interpretability Alignmentarxiv.org
• Why Frontier AI Labs Fight to Hide Chain of Thought — Ilia Shumailov & Alexander Panfilovyoutube.com
• When Chain of Thought is Necessary, Language Models Struggle to Evade Monitorsarxiv.org
• Assessing skeptical views of interpretability research | Christopher Pottsweb.stanford.edu
• the case for CoT unfaithfulness is overstated — LessWronglesswrong.com
• [2603.05706] Reasoning Models Struggle to Control their Chains of Thoughtarxiv.org
• Transformers Provably Learn to Internalize Chain-of-Thoughtarxiv.org
• Don’t Outsource Your Thinkingteltam.github.io
• Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscationarxiv.org
• Policy Options for Preserving Chain of Thought Monitorability — Institute for AI Policy and Strategyiaps.ai