mlsanityData to 7 Oct 2026

Reasoning Theater: Disentangling Model Beliefs from Chain-of-Thought

arXiv 2603.05488 ↗First posted 5 Mar 2026cs.CL cs.AI cs.LGcited by 35

On the map: LLM reasoning → Prompting and chain-of-thought reasoning

Abstract (arXiv)

We provide evidence of performative chain-of-thought (CoT) in reasoning models, where a model becomes strongly confident in its final answer, but continues generating tokens without revealing its internal belief. Our analysis compares activation probing, early forced answering, and a CoT monitor across two large models (DeepSeek-R1 671B & GPT-OSS 120B) and find task difficulty-specific differences: The model's final answer is decodable from activations far earlier in CoT than a monitor is able to say, especially for easy recall-based MMLU questions. We contrast this with genuine reasoning in difficult multihop GPQA-Diamond questions. Despite this, inflection points (e.g., backtracking, 'aha' moments) occur almost exclusively in responses where probes show large belief shifts, suggesting these behaviors track genuine uncertainty rather than learned "reasoning theater." Finally, probe-guided early exit reduces tokens by up to 80% on MMLU and 30% on GPQA-Diamond with similar accuracy, positioning attention probing as an efficient tool for detecting performative reasoning and enabling adaptive computation.

Key facts. Cited by 35 papers in the map (data through 7 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 29 of its 47 references that are in the map. First posted on arXiv 5 Mar 2026.

Builds on

29 of 47 references.

Closest ancestors · 13

Same direction, newest first

  1. 2026A Positive Case for Faithfulness: LLM Self-Explanations Help Predict Model Behavior2 Feb
  2. 2025Are language models aware of the road not taken? Token-level uncertainty and hidden state dynamics6 Nov
  3. 2025Thought Branches: Interpreting LLM Reasoning Requires Resampling31 Oct
  4. 2025Is It Thinking or Cheating? Detecting Implicit Reward Hacking by Measuring Reasoning Effort1 Oct
  5. 2025Can We Predict Alignment Before Models Finish Thinking? Towards Monitoring Misaligned Reasoning Models16 Jul
  6. 2025Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety15 Jul
  7. 2025When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors7 Jul
  8. 2025Thought Anchors: Which LLM Reasoning Steps Matter?23 Jun
  9. 2025Reasoning Models Don't Always Say What They Think8 May
  10. 2025Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation14 Mar
  11. 2023Measuring Faithfulness in Chain-of-Thought Reasoning17 Jul
  12. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting7 May
1 more
  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Models28 Jan

Other references · 16

  1. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  2. 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,655
  3. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  4. 2022Language Models (Mostly) Know What They Knowcited by 1,783
  5. 2016Understanding intermediate layers using linear classifier probescited by 1,279
  6. 2025gpt-oss-120b & gpt-oss-20b Model Cardcited by 1,258
  7. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,130
  8. 2021Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigmcited by 599
  9. 2024Transcoders Find Interpretable LLM Feature Circuitscited by 186
  10. 2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verificationcited by 154
  11. 2024Faithfulness vs. Plausibility: On the (Un)Reliability of Explanations from Large Language Modelscited by 104
  12. 2025Are DeepSeek R1 And Other Reasoning Models More Faithful?cited by 69
4 more
  1. 2025Understanding Reasoning in Thinking Language Models via Steering Vectorscited by 69
  2. 2025Detecting High-Stakes Interactions with Activation Probescited by 45
  3. 2025Base Models Know How to Reason, Thinking Models Learn Whencited by 25
  4. 2025Reasoning-Finetuning Repurposes Latent Representations in Base Modelscited by 23

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 35.

35 papers.

Sort:

By area:

  1. 2026The Hidden Signal of Verifier Strictness: Controlling and Improving Step-Wise Verification via Selective Latent Steeringcited by 5
  2. 2026Not All LLM Reasoning is Visible in the Chain-of-Thoughtcited by 5
  3. 2026Why Models Know But Don't Say: Chain-of-Thought Faithfulness Divergence Between Thinking Tokens and Answers in Open-Weight Reasoning Modelscited by 4
  4. 2026Beyond the Commitment Boundary: Probing Epiphenomenal Chain-of-Thought in Large Reasoning Modelscited by 4
  5. 2026Reliable Chain-of-Thought via Prefix Consistencycited by 3
  6. 2026Faithfulness Metrics Don't Measure Faithfulness: A Meta-Evaluation with Ground Truthcited by 3
  7. 2026Observable Patterns Are Not Explanations: A Causal-Geometric Analysis of Latent Reasoning Modelscited by 3
  8. 2026Do Thinking Tokens Help with Safety?cited by 3
  9. 2026Your Agentic LLMs Secretly Encode Indirect Prompt-Injection Exposure in Hidden Statescited by 3
  10. 2026Therefore I am. I Thinkcited by 2
  11. 2026The Detection-Extraction Gap: Models Know the Answer Before They Can Say Itcited by 2
  12. 2026When Does a Language Model Commit? A Finite-Answer Theory of Pre-Verbalization Commitmentcited by 2
23 more
  1. 2026When Reasoning Traces Become Performative: Step-Level Evidence that Chain-of-Thought Is an Imperfect Oversight Channelcited by 2
  2. 2026Agent trajectories as programs: fingerprinting and programming coding-agent behaviorcited by 2
  3. 2026What LLM Forecasters Know but Don't Say: Probing Internal Representations for Calibration and Faithfulnesscited by 2
  4. 2026Answer First, Reason Later: When Commitment Order Costs Accuracy in Diffusion Language Modelscited by 2
  5. 2026Persona Non Grata: Single-Method Safety Evaluation Is Incomplete for Persona-Imbued LLMscited by 1
  6. 2026Reasoning Models Know What's Important, and Encode It in Their Activationscited by 1
  7. 2026CopT: Contrastive On-Policy Thinking with Continuous Spaces for General and Agentic Reasoningcited by 1
  8. 2026How Language Models Fail: Token-Level Signatures of Committed and Persistent Reasoning Failurescited by 1
  9. 2026How Much Does a Reasoning Summary Reveal? An Observability Ladder for Large Language Modelscited by 1
  10. 2026Forking Fast: Efficiently Estimating Uncertainty Dynamics in Text Generationcited by 1
  11. 2026Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluationscited by 1
  12. 2026Internalized Reasoning for Long-Context Visual Document Understandingcited by 0
  13. 2026LLM Reasoning Is Latent, Not the Chain of Thoughtcited by 0
  14. 2026The Cognitive Penalty: Ablating System 1 and System 2 Reasoning in Edge-Native SLMs for Decentralized Consensuscited by 0
  15. 2026Forecasting Future Behavior as a Learning Taskcited by 0
  16. 2026Committed Before Reasoning: Behavioral Reproduction and Preliminary Activation-Level Evidence of Answer Pre-Commitment in an Open-Weight LLMcited by 0
  17. 2026On the Generalization of Steering Vectors for Chain-of-Thought Faithfulnesscited by 0
  18. 2026Amplified Does Not Mean Predictive: Reasoning Behaviors in Thinking Modelscited by 0
  19. 2026Bayesian Intelligence from the Outsidecited by 0
  20. 2026GYROval: A Robust Benchmark for Cultural Value Orientation in Large Language Modelscited by 0
  21. 2026When Confidence Rises Too Early: Detecting Shortcut Reasoning via Premature Answer Commitmentcited by 0
  22. 2026Rethinking Reasoning Paths as Phase-Structured Trajectoriescited by 0
  23. 2026On Language Drift during RLVR Post-Trainingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Deep Hidden Cognition Facilitates Reliable Chain-of-Thought Reasoning
  2. 2026When Chain-of-Thought Fails, the Solution Hides in the Hidden States
  3. 2026Emergent Search and Backtracking in Latent Reasoning Models
  4. 2026Post-Hoc Reasoning in Chain of Thought: Decoding and Steering Pre-Committed Answers
  5. 2026Does Out-of-Sight Equal Out-of-Mind in CoT Monitorability?
  6. 2026Drop the Act: Probe-Filtered RL for Faithful Chain-of-Thought Reasoning
  7. 2026Do Latent-CoT Models Think Step-by-Step? A Mechanistic Study on Sequential Reasoning Tasks
  8. 2025How does Chain of Thought Think? Mechanistic Interpretability of Chain-of-Thought Reasoning with Sparse Autoencoding