CoT Monitoring

CoT: Chain-of-Thought

Chain-of-thought monitoring asks whether a model’s visible reasoning reveals the factors behind its answers and actions, so people or automated monitors can detect errors, unsafe behavior, and deception. Researchers test this with benchmarks and interventions such as hint changes, counterfactuals, and adversarial prompts, and build monitors using reasoning text, model activations, gradients, or action signals. Training approaches try to make reasoning more faithful or legible through targeted rewards, behavior cues, and monitor-focused fine-tuning. A central tension is that traces can omit influential information or be deliberately shaped to evade oversight, with monitorability varying by task, language, and training method.

Latest papers

40 papers11 papers in the last four weeks, up 175% on the four weeks before. 0.1% of all new papers.

Learning to Be Monitored: Causal Bounds and Intervention-Aware Post-Training Recipes for Chain-of-Thought Monitorability
Oct 22, 2026Qiang Fu, Xinyue Li, Andre Wibisono +1CoT Monitoring LLM Safety Monitoring
Yale University
Monitorability Is Trainable: Post-Training for Chain-of-Thought Monitorability
Oct 21, 2026Usman Anwar, Tim Bakker, Christos Louizos +1CoT Monitoring LLM Post-Training
University of Cambridge · Qualcomm AI Research · Mila, University of Montreal
Reward Misspecification Degrades Chain-of-Thought Monitorability
Oct 20, 2026Tianyi Lorena Yan, John HewittLLM Safety Monitoring CoT Monitoring
Columbia University
CoTs remain monitorable when being unfaithful requires computation: Replicating and Extending the Necessity Effect Across Eleven Models
Oct 17, 2026Arav Dhoot, Yixiong Hao, Zephaniah RoeFaithfulness of Language Model Reasoning Faithfulness of Language Model Explanations
Columbia University · Second Look Research
OwlSight: A Panoramic Monitor for Chain-of-Thought Reasoning
Oct 16, 2026Lewen Yan, Yanxu Zhu, Jie Zhang +1CoT Monitoring LLM Safety Monitoring
Reasoning Models Can Control Their Chains of Thought When They Tell Themselves To
Oct 16, 2026Uriel Dolev, Idan Schwartz, Yoav GoldbergLanguage Model Reasoning Evaluation LLM Prompting
Bar-Ilan University · Allen Institute for AI
Chain of Thought Monitoring Can Fail Even When CoT Is Necessary
Oct 13, 2026Lennart Bürger, Piotr Komorowski, Moritz Weckbecker +4CoT Monitoring LLM Auditing
Fraunhofer HHI · TU Dublin · TU Berlin +1
Reasoning-Token Spikes Under Prompted Untruthful Responses in Large Language Models
Oct 7, 2026Maverick Morales, Tomáš Dominik, Vermut Gao +4Deception in Language Models LLM Auditing
Chapman University, CA, USA · INSERM U992, Cognitive Neuroimaging Unit, NeuroSpin Center, Gif sur Yvette 91191, France · University of California, Los Angeles, CA, USA +1
On the Chain-of-Thought Monitorability of Looped Language Models
Oct 2, 2026Han Wang, Ishwar B Balappanawar, Huan ZhangCoT Monitoring Recursive Language Models
University of Illinois Urbana-Champaign
On Language Drift during RLVR Post-Training
Oct 1, 2026Michael Sullivan, Alexander KollerRL for Language Model Reasoning Reinforcement Learning with Verifiable Rewards
Department of Language Science and Technology, Saarland University, Saarbrücken, Germany.
Hidden Reasoning Must Leak, but Need Not Be Readable: Fundamental Opportunities and Limits for Chain-of-Thought Monitoring
Sep 29, 2026Mohammadali Mohammadkhani, Madhava Krishna, Yash Sarrof +1CoT Monitoring Large Reasoning Models
Saarland University, Saarland Informatics Campus · Zuse School ELIZA
Monitor Jailbreaking: Evading Chain-of-Thought Monitoring Without Encoded Reasoning
Sep 25, 2026Julian SchulzCoT Monitoring Language Model Safety Evaluation
Meridian Cambridge
From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought
Sep 21, 2026Renee Jia, Di MuCoT Faithfulness Causal Interventions in Language Models
R2M AI, Cornell University · R2M AI, Carnegie Mellon University
Corrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection
Sep 14, 2026Keertana Chidambaram, Andrew Ilyas, Vasilis SyrgkanisLLM Security Prompt Injection Attacks on AI Agents
Stanford University · CMU
The Answer Is Not the Argument
Aug 31, 2026Will Yeadon, Sergio Juárez, Paul Mackay +5LLM Answer Verification CoT Monitoring
Department of Physics, Durham University, Durham DH1 3LE, UK · Escuela de Ingenier´ıa de Telecomunicaci´on, Department of Signal Theory and Communications, University of Vigo, Vigo E-36310, Spain · Independent Researcher
INTENT-AS-A-TOOL Makes it Easy to Track Agentic Misalignment
Aug 27, 2026Yutong Zhang, Jianshuo Dong, Peng Xu +5AI Agent Monitoring LLM Agent Safety
Tsinghua University · MatrixOrigin · SiliconProspect AI +1
Visible Reasoning and Indirect Prompt-Injection Monitorability Across English, Tamil, and Tanglish
Aug 15, 2026Madhusudhanan GLanguage Model Safety Evaluation CoT Monitoring
Vellore Institute of Technology, India
Does Out-of-Sight Equal Out-of-Mind in CoT Monitorability?
Aug 5, 2026Pedro Ferreira, Wilker Aziz, Ivan TitovCoT Monitoring Latent CoT Reasoning
University of Amsterdam · University of Edinburgh
Chain-of-Thought Monitoring Can Be Unreliable in Implicit-Influence Settings
Aug 5, 2026Agatha Duzan, Asa Cooper SticklandLLM Safety Benchmarks CoT Monitoring
EPFL · UK AI Security Institute
Evading Chain-of-Thought Monitoring Through Model Poisoning
Aug 3, 2026Giorgio Severi, Shujaat Mirza, Blake Bullwinkel +1LLM Backdoor Attacks LLM Auditing
Microsoft
A False Average: Pooled CoT-Monitor Accuracy Conceals a Reasoning-Dependent Fragility
Aug 1, 2026Shikhar Shiromani, Leo Richter, Noah SiegelReward Hacking Adversarial Attacks on LLMs
Pivotal Research · University College London
Persuasion Attacks Can Decrease Effectiveness of CoT Monitoring
Jul 9, 2026Jennifer Za, Julija Bainiaksina, Nikita Ostrovsky +2Language Model Safety Evaluation Adversarial Attacks on LLMs
LASR Labs · University College London · Google DeepMind
Length Penalties Make Chain-of-Thought Less Monitorable
Jul 8, 2026Bryce LittleCoT Faithfulness Overthinking in Language Models
Can Reasoning Models Detect Changes to their Chains of Thought?
Jun 20, 2026Sathvik Napa, Utkarsh Singh, Chengyuan Xue +2Causal Interventions in Language Models Metacognition in Language Models
Johns Hopkins University
Bootstrapped Monitoring: Leveraging Transparent Reasoning to Oversee Stronger AI Agents
Jun 10, 2026Frank Xiao, Mary PhuongAI Agent Safety AI Agent Monitoring
California Institute of Technology
The Fragility of Chain-of-Thought Monitoring Across Typologically Diverse Languages
May 27, 2026Eric Onyame, Runtao Zhou, Kowshik Thopalli +2CoT Faithfulness Multilingual Language Model Evaluation
University of Virginia · Lawrence Livermore National Laboratory
Training on Documents About Monitoring Leads to CoT Obfuscation
May 14, 2026Reilly Haskins, Bilal Chughtai, Joshua EngelsCoT Monitoring
University of Canterbury · Google DeepMind · Pivotal Research
CoT-Guard: Small Models for Strong Monitoring
May 12, 2026Nirav Diwan, Han Wang, Berkcan Kapusuzoglu +6LLM Auditing Adversarial Attacks on LLMs
†University of Illinois Urbana-Champaign · ♢Capital One
Behavior Cue Reasoning: Monitorable Reasoning Improves Efficiency and Safety through Oversight
May 7, 2026Christopher Z. Cui, Taylor W. Killian, Prithviraj AmmanabroluScalable Oversight LLM Safety
University of California, San Diego · 2Brigham Young University
Detecting and Suppressing Reward Hacking with Gradient Fingerprints
Apr 17, 2026Songtao Wang, Quang Hieu Pham, Fangcong Yin +4Reward Hacking Reinforcement Learning with Verifiable Rewards
AUniversity of Alberta · N New York University · LLMU Munich +1

Oct 22, 2026J/K move · Enter open · S save

Learning to Be Monitored: Causal Bounds and Intervention-Aware Post-Training Recipes for Chain-of-Thought Monitorability

Qiang Fu, Xinyue Li, Andre Wibisono, Arman Cohan

Yale University

CoT Monitoring LLM Safety Monitoring Causal Interventions in Language Models

Open paper AskPDF arXiv

Chain-of-thought (CoT) monitoring aims to detect potentially harmful behaviors in large language models (LLMs) from their CoTs. Intervention-based evaluations measure this property by asking whether a monitor can detect the role of an intervention from the CoT when the targeted behavior occurs, typically using minimal-criterion sensitivity (

\mathsf{TPR_{mc}}

) proposed by~\citep{guan2025monitoring}. However, a target behavior observed under intervention is not necessarily caused by it, and we show that this metric can overestimate how often a monitor detects target behaviors actually caused by the intervention. Specifically, we establish

\mathsf{TPR_{mc}}

as an upper bound on this causal detection rate and derive a complementary lower bound, which provides a tighter evaluation framework for CoT monitorability. With this framework, we study whether LLMs can be trained to make their CoTs easier to monitor. We first construct a training dataset of paired control--intervention tasks spanning multiple intervention types. Using the dataset, we design and apply three intervention-specific training methods based on off-policy self-distillation, reinforcement learning, and on-policy self-distillation to train separate variants of each base model. Across various model-monitor configurations, all three methods generally improve CoT monitorability over the corresponding base models under our extended evaluation framework. Our code is available at https://github.com/xinyue-lxy/learning-to-be-monitored.