Chain-of-thought (CoT) monitoring is a promising tool for detecting misbehaviors and understanding the motivations of modern reasoning models. However, if models can control what they verbalize in their CoT, it could undermine CoT monitorability. To measure this undesirable capability -- CoT controllability -- we introduce the CoT-Control evaluation suite, which includes tasks that require models to solve problems while adhering to CoT instructions, e.g., reasoning about a genetics question without using the word 'chromosome'. We show that reasoning models possess significantly lower CoT controllability than output controllability; for instance, Claude Sonnet 4.5 can control its CoT only 2.7% of the time but 61.9% when controlling its final output. We also find that CoT controllability is higher for larger models and decreases with more RL training, test-time compute, and increased problem difficulty. CoT controllability failures extend even to situations in which models are given incentives (as opposed to direct requests) to evade CoT monitors, although models exhibit slightly higher controllability when they are told they are being monitored. Similarly, eliciting controllability by adversarially optimizing prompts does not meaningfully increase controllability. Our results leave us cautiously optimistic that CoT controllability is currently unlikely to be a failure mode of CoT monitorability. However, the mechanism behind low controllability is not well understood. Given its importance for maintaining CoT monitorability, we recommend that frontier labs track CoT controllability in future models.
Does Chain-of-Thought Reasoning Really Reduce Harmfulness from Jailbreaking?
IIIS, Tsinghua University(清华大学信息学院);Shanghai Qi Zhi Institute(上海启智研究院);Department of Statistics and Data Science, University of Pennsylvania(宾夕法尼亚大学统计与数据科学系);Data Agent Group, Beijing Academy of Artificial Intelligence(北京人工智能研究院数据代理组)
Streaming Hallucination Detection in Long Chain-of-Thought Reasoning
Beijing University of Posts and Telecommunications(北京邮电大学);Nanyang Technological University(南洋理工大学);Southwest Jiaotong University(西南交通大学);Renmin University of China(中国人民大学)
Reason from Future: Reverse Thought Chain Enhances LLM Reasoning
College of Computer Science and Technology, Zhejiang University(浙江大学计算机科学与技术学院);Alibaba Group(阿里巴巴集团);State Key Laboratory of Transvascular Implantation Devices and TIDRI(血管植入设备国家重点实验室);Liangzhu Laboratory and WeDoctor Cloud(良渚实验室和WeDoctor云);Zhejiang Key Laboratory of Medical Imaging Artificial Intelligence(浙江医学影像人工智能重点实验室)
Heuristic Classification of Thoughts Prompting (HCoT): Integrating Expert System Heuristics for Structured Reasoning into Large Language Models
Computer Network Information Center, Chinese Academy of Sciences(中国科学院计算机网络信息中心);Shenyang Aerospace University(沈阳航空航天大学);ZGC LAB, Beijing,China(北京ZGC实验室);Institute for Network Sciences and Cyberspace, Tsinghua University(清华大学网络科学与网络空间研究院)
ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning
Shanghai Artificial Intelligence Laboratory(上海人工智能实验室);University of Science and Technology of China(中国科学技术大学);MMLab, The Chinese University of Hong Kong(香港中文大学MMLab)
Retrieval-of-Thought: Efficient Reasoning via Reusing Thoughts
Department of Computer Science and Engineering, University of Minnesota(明尼苏达大学计算机科学与工程系);Department of Computer Science and Engineering, University of California Santa Cruz(加州大学圣克鲁兹分校计算机科学与工程系);Argonne National Laboratory(阿贡国家实验室)
Training Multimodal Large Reasoning Models Needs Better Thoughts: A Three-Stage Framework for Long Chain-of-Thought Synthesis and Selection
School of Computer Science and Engineering(计算机科学与工程学院);Key Laboratory of Computer Network and Information Integration (SEU), Ministry of Education(计算机网络与信息集成重点实验室(SEU))
ThoughtSource: A central hub for large language model reasoning data
Medical University of Vienna(维也纳医科大学);Technical University of Denmark(丹麦技术大学);Copenhagen University Hospital(哥本哈根大学医院);The Royal Institute of Technology (KTH)(皇家理工学院(KTH))
Not All Thoughts are Generated Equal: Efficient LLM Reasoning via Multi-Turn Reinforcement Learning
AI Thrust, The Hong Kong University of Science and Technology (Guangzhou)(香港科学与技术大学(广州)人工智能方向);Didichuxing Co. Ltd(滴滴出行有限公司);CSE, The Hong Kong University of Science and Technology(香港科学与技术大学计算机科学与工程系)
Chaining Simultaneous Thoughts for Numerical Reasoning
Tsinghua University(清华大学);Institute for Artificial Intelligence(人工智能研究院);State Key Lab of Intelligent Technology and Systems(智能技术与系统国家重点实验室);Beijing National Research Center for Information Science and Technology(北京信息科学与技术国家研究中心)