Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability

N/A

Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability Recent findings suggest that misaligned models may exhibit deceptive behavior, raising concerns about output trustworthiness. Chain-of-thought (CoT) is a promising tool for alignment monitoring: when models articulate their reasoning faithfully, monitors can detect and mitigate harmful behaviors before undesirable outcomes occur. However, a key uncertainty is: Can models obfuscate their CoT in order to pursue hidden adversarial objectives while evading detection? To answer this question and thus stress-te

saved by

• Yeedrag

related reading

• When Chain of Thought is Necessary, Language Models Struggle to Evade Monitorsarxiv.org
• Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscationarxiv.org
• [2603.05706] Reasoning Models Struggle to Control their Chains of Thoughtarxiv.org
• Externalized reasoning oversight: a research direction for language model alignment — AI Alignment Forumalignmentforum.org
• Reasoning models don't always say what they think \ Anthropicanthropic.com
• What’s your AI thinking? - AI Digesttheaidigest.org
• [2505.05410] Reasoning Models Don't Always Say What They Thinkarxiv.org
• How AI Is Learning to Think in Secret — LessWronglesswrong.com
• The fragile foundations of CoT monitoring | Christopher Pottsweb.stanford.edu
• [2510.09714] All Code, No Thought: Current Language Models Struggle to Reason in Ciphered Languagearxiv.org
• [2512.18311] Monitoring Monitorabilityarxiv.org
• [2510.27338] Reasoning Models Sometimes Output Illegible Chains of Thoughtarxiv.org