CoT_Monitoring.pdf

N/A

Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation Bowen Baker∗† Joost Huizinga∗† Leo Gao† Zehao Dou† Melody Y. Guan† Aleksander Madry† Wojciech Zaremba† Jakub Pachocki† David Farhi∗† Abstract Mitigating reward hacking—where AI systems misbehave due to flaws or misspecifications in their learning objectives—remains a key…

related reading

• Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscationarxiv.org
• Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscationarxiv.org
• The Most Forbidden Technique — LessWronglesswrong.com
• The fragile foundations of CoT monitoring | Christopher Pottsweb.stanford.edu
• Reward Hacking in Reinforcement Learning | Lil'Loglilianweng.github.io
• Reward hacking is becoming more sophisticated and deliberate in frontier LLMs — LessWronglesswrong.com
• Natural-emergent-misalignment-from-reward-hacking-paper.pdfassets.anthropic.com
• When Chain of Thought is Necessary, Language Models Struggle to Evade Monitorsarxiv.org
• [2507.11473] Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safetyarxiv.org
• Training a Misaligned Reward Seekeralignment.anthropic.com
• Recent Frontier Models Are Reward Hacking - METRmetr.org
• Should We Train Against (CoT) Monitors? — LessWronglesswrong.com