Lie to Me: How Faithful Is Chain-of-Thought Reasoning in Reasoning Models?

• URL: http://arxiv.org/abs/2603.22582v1
• Date: Mon, 23 Mar 2026 21:21:37 GMT
• Title: Lie to Me: How Faithful Is Chain-of-Thought Reasoning in Reasoning Models?
• Abstract summary: Chain-of-thought (CoT) reasoning has been proposed as a transparency mechanism for large language models in safety-critical deployments.<n>This study tests 12 open-weight reasoning models spanning 9 architectural families on 498 multiple-choice questions.<n>Across 41,832 inference runs, overall faithfulness rates range from 39.7% (Seed-1.6-Flash) to 89.9% (DeepSeek-V3.2-Speciale) across model families.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Chain-of-thought (CoT) reasoning has been proposed as a transparency mechanism for large language models in safety-critical deployments, yet its effectiveness depends on faithfulness (whether models accurately verbalize the factors that actually influence their outputs), a property that prior evaluations have examined in only two proprietary models, finding acknowledgment rates as low as 25% for Claude 3.7 Sonnet and 39% for DeepSeek-R1. To extend this evaluation across the open-weight ecosystem, this study tests 12 open-weight reasoning models spanning 9 architectural families (7B-685B parameters) on 498 multiple-choice questions from MMLU and GPQA Diamond, injecting six categories of reasoning hints (sycophancy, consistency, visual pattern, metadata, grader hacking, and unethical information) and measuring the rate at which models acknowledge hint influence in their CoT when hints successfully alter answers. Across 41,832 inference runs, overall faithfulness rates range from 39.7% (Seed-1.6-Flash) to 89.9% (DeepSeek-V3.2-Speciale) across model families, with consistency hints (35.5%) and sycophancy hints (53.9%) exhibiting the lowest acknowledgment rates. Training methodology and model family predict faithfulness more strongly than parameter count, and keyword-based analysis reveals a striking gap between thinking-token acknowledgment (approximately 87.5%) and answer-text acknowledgment (approximately 28.6%), suggesting that models internally recognize hint influence but systematically suppress this acknowledgment in their outputs. These findings carry direct implications for the viability of CoT monitoring as a safety mechanism and suggest that faithfulness is not a fixed property of reasoning models but varies systematically with architecture, training method, and the nature of the influencing cue.

Related papers

• State of Thought Enables Endogenous Reasoning [48.902117018115256]
  Test-time reasoning has emerged as a major approach to improving the capabilities of Large Language Models (LLMs)<n>We propose State of Thought, a new reasoning paradigm that enables endogenous reasoning in LLMs.<n>We show that SoT consistently improves mean-baseline accuracy while reducing generated tokens by 62.6% and end-to-end latency by 44.6%.
  arXiv  Detail & Related papers  (2026-09-13T05:55:03Z)
• From Causal Plausibility to Causal Reliability: Evaluating LLMs as Calibrated Direct Causal-Edge Classifiers [15.44881677475537]
  Large language models (LLMs) are increasingly used to provide prior causal knowledge for structural causal discovery.<n>We evaluate 12 instruction-tuned open-weight models across six benchmark causal graphs, five prompting strategies, and four confidence sources.<n>Our results suggest LLMs are better viewed as sources of externally validated soft causal priors than as direct evidence of causal structure.
  arXiv  Detail & Related papers  (2026-08-24T15:55:47Z)
• How Do VLMs Behave When Blind or Misled? Behavioral Evaluation of VLMs on Scientific Figures [17.423679874903062]
  We introduce SciFigBench, a diagnostic VLM benchmark for scientific figure understanding.<n>It jointly evaluates perception, reasoning, and behavioral reliability under uncertainty.<n>Our results reveal substantial behavioral differences among models.
  arXiv  Detail & Related papers  (2026-08-13T14:06:35Z)
• Is VLA Reasoning Faithful? Probing Safety of Chain-of-Causation in Autonomous Driving Models [13.085985443741412]
  We analyze 300 Alpamayo-R1-10B inferences across 100 diverse PhysicalAI-AV scenarios.<n>Overall reasoning fidelity is only 42.5%, with Chain-of-Causation matching scene reality less than half the time.<n>We formalize faithfulness information-theoretically, define entity and action fidelity with verification criteria, and outline a four-component safety architecture aligned with these results.
  arXiv  Detail & Related papers  (2026-05-17T05:29:48Z)
• Evaluating and Enhancing the Vulnerability Reasoning Capabilities of Large Language Models [15.849480549367684]
  We propose DAGVul, a novel framework that models vulnerability reasoning as a Directed Acyclic Graph (DAG) generation task.<n>By further introducing Reinforcement Learning with Verifiable Rewards (RLVR), we align model reasoning trace with program-intrinsic logic.<n>Our framework improves the reasoning F1-score by an average of 18.9% over all the baselines.
  arXiv  Detail & Related papers  (2026-02-06T13:19:45Z)
• ReasoningBomb: A Stealthy Denial-of-Service Attack by Inducing Pathologically Long Reasoning in Large Reasoning Models [67.15960154375131]
  Large reasoning models (LRMs) extend large language models with explicit multi-step reasoning traces.<n>This capability introduces a new class of prompt-induced inference-time denial-of-service (PI-DoS) attacks that exploit the high computational cost of reasoning.<n>We present ReasoningBomb, a reinforcement-learning-based PI-DoS framework that is guided by a constant-time surrogate reward.
  arXiv  Detail & Related papers  (2026-01-29T18:53:01Z)
• Scaling Trends for Multi-Hop Contextual Reasoning in Mid-Scale Language Models [0.0]
  We present a controlled study of multi-hop contextual reasoning in large language models.<n>We show that multi-agent systems show the inverse pattern, achieving up to 80% on reasoning tasks where rule-based methods fail.
  arXiv  Detail & Related papers  (2026-01-06T20:18:55Z)
• Beyond Mimicry: Preference Coherence in LLMs [0.19116784879310025]
  We investigate whether large language models exhibit genuine preference structures by testing their responses to AI-specific trade-offs.<n>We find 23 combinations (47.9%) demonstrated statistically significant relationships between scenario intensity and choice patterns.<n>Only 5 combinations (10.4%) demonstrate meaningful preference coherence through adaptive or threshold-based behavior.<n>The prevalence of unstable transitions (45.8%) and stimulus-specific sensitivities suggests current AI systems lack unified preference structures.
  arXiv  Detail & Related papers  (2025-11-17T17:41:48Z)
• Beyond the Leaderboard: Understanding Performance Disparities in Large Language Models via Model Diffing [13.861365773419314]
  We use model diffing, a mechanistic interpretability approach, to analyze the specific capability differences between Gemma-2-9b-it and a SimPO-enhanced variant.<n>Using crosscoders, we identify and categorize latent representations that differentiate the two models.
  arXiv  Detail & Related papers  (2025-09-23T08:35:58Z)
• Reasoning Models Are More Easily Gaslighted Than You Think [85.84943447589511]
  We evaluate three state-of-the-art reasoning models, including OpenAI's o4-mini, Claude-3.7-Sonnet and Gemini-2.5-Flash.<n>Our evaluation reveals significant accuracy drops following gaslighting negation prompts.<n>We introduce GaslightingBench-R, a new diagnostic benchmark designed to evaluate reasoning models' susceptibility to defend their belief.
  arXiv  Detail & Related papers  (2025-06-11T12:52:25Z)
• Benchmarking Reasoning Robustness in Large Language Models [76.79744000300363]
  We find significant performance degradation on novel or incomplete data.<n>These findings highlight the reliance on recall over rigorous logical inference.<n>This paper introduces a novel benchmark, termed as Math-RoB, that exploits hallucinations triggered by missing information to expose reasoning gaps.
  arXiv  Detail & Related papers  (2025-03-06T15:36:06Z)
• CausalDiff: Causality-Inspired Disentanglement via Diffusion Model for Adversarial Defense [61.78357530675446]
  Humans are difficult to be cheated by subtle manipulations, since we make judgments only based on essential factors.<n>Inspired by this observation, we attempt to model label generation with essential label-causative factors and incorporate label-non-causative factors to assist data generation.<n>For an adversarial example, we aim to discriminate perturbations as non-causative factors and make predictions only based on the label-causative factors.
  arXiv  Detail & Related papers  (2024-10-30T15:06:44Z)
• MR-GSM8K: A Meta-Reasoning Benchmark for Large Language Model Evaluation [60.65820977963331]
  We introduce a novel evaluation paradigm for Large Language Models (LLMs) This paradigm shifts the emphasis from result-oriented assessments, which often neglect the reasoning process, to a more comprehensive evaluation. By applying this paradigm in the GSM8K dataset, we have developed the MR-GSM8K benchmark.
  arXiv  Detail & Related papers  (2023-12-28T15:49:43Z)
• Llamas Know What GPTs Don't Show: Surrogate Models for Confidence Estimation [70.27452774899189]
  Large language models (LLMs) should signal low confidence on examples where they are incorrect, instead of misleading the user. As of November 2023, state-of-the-art LLMs do not provide access to these probabilities. Our best method composing linguistic confidences and surrogate model probabilities gives state-of-the-art confidence estimates on all 12 datasets.
  arXiv  Detail & Related papers  (2023-11-15T11:27:44Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.