mlsanityData to 14 Oct 2026

Decoding Hidden Deception in Reasoning LLMs: Activation Explainers for Deception Auditing

arXiv 2606.17478 ↗First posted 16 Jun 2026cs.CL cs.AIcited by 2

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

Abstract (arXiv)

As LLMs acquire stronger reasoning capabilities, deceptive behavior becomes an increasingly serious safety concern. Existing deception monitors either score visible transcripts or derive scalar probe scores from representation vectors, leaving little inspectable evidence about why a response is suspicious. We introduce STATEWITNESS, an activation explainer for deception auditing. A separate decoder reads a target model's hidden states, then answers natural-language queries or emits structured reports about them. We evaluate STATEWITNESS on two target reasoning LLMs across seven deception datasets. STATEWITNESS reaches 0.916 mean AUROC, a relative gain of 11.6% over the best black-box text monitor and 25.0% over the best activation-probe baseline under the same evaluation protocol. When combined with existing monitors, STATEWITNESS reduces missed deceptive examples in simple threshold ensembles. Beyond scalar detection, the decoder returns query-level answers, schema reports, and token- or sentence-level evidence traces for human inspection. We view this interface as a potential building block for broader interpretability and alignment tools.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 46 of its 48 references that are in the map. First posted on arXiv 16 Jun 2026.

Builds on

46 of 48 references.

Closest ancestors · 8

Same direction, newest first

  1. 2025Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety15 Jul
  2. 2025When Thinking LLMs Lie: Unveiling the Strategic Deception in Representations of Reasoning Models5 Jun
  3. 2025Mitigating Deceptive Alignment via Self-Monitoring24 May
  4. 2025Reasoning Models Don't Always Say What They Think8 May
  5. 2025Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation14 Mar
  6. 2025Measuring Chain of Thought Faithfulness by Unlearning Reasoning Steps20 Feb
  7. 2024Chain-of-Thought Unfaithfulness as Disguised Accuracy22 Feb
  8. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting7 May

Other references · 38

  1. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  2. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  3. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  4. 2023Towards Understanding Sycophancy in Language Modelscited by 1,040
  5. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 712
  6. 2024Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingcited by 529
  7. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 509
  8. 2024Frontier Models are Capable of In-context Schemingcited by 287
  9. 2019Risks from Learned Optimization in Advanced Machine Learning Systemscited by 258
  10. 2022The Alignment Problem from a Deep Learning Perspectivecited by 258
  11. 2023AI Control: Improving Safety Despite Intentional Subversioncited by 229
  12. 2023Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmarkcited by 182
26 more
  1. 2023Large Language Models can Strategically Deceive their Users when Put Under Pressurecited by 146
  2. 2023Scheming AIs: Will AIs fake alignment during training in order to get power?cited by 97
  3. 2023How to Catch an AI Liar: Lie Detection in Black-Box LLMs by Asking Unrelated Questionscited by 83
  4. 2025Are DeepSeek R1 And Other Reasoning Models More Faithful?cited by 72
  5. 2025The MASK Benchmark: Disentangling Honesty From Accuracy in AI Systemscited by 72
  6. 2025SHADE-Arena: Evaluating Sabotage and Monitoring in LLM Agentscited by 71
  7. 2025Detecting Strategic Deception Using Linear Probescited by 68
  8. 2025Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainerscited by 52
  9. 2024LatentQA: Teaching LLMs to Decode Activations Into Natural Languagecited by 51
  10. 2024Uncovering Deceptive Tendencies in Language Models: A Simulated Company AI Assistantcited by 38
  11. 2026AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviorscited by 38
  12. 2024Towards evaluations-based safety cases for AI schemingcited by 33
  13. 2025Evaluating the Paperclip Maximizer: Are RL-Based Language Models More Likely to Pursue Instrumental Goals?cited by 33
  14. 2025Training Language Models to Explain Their Own Computationscited by 30
  15. 2025DarkBench: Benchmarking Dark Patterns in Large Language Modelscited by 28
  16. 2025Eliciting Secret Knowledge from Language Modelscited by 27
  17. 2025AgentMisalignment: Measuring the Propensity for Misaligned Behaviour in LLM-Based Agentscited by 22
  18. 2025Liars' Bench: Evaluating Lie Detectors for Language Modelscited by 22
  19. 2025Training LLMs for Honesty via Confessionscited by 22
  20. 2025Why Do Some Language Models Fake Alignment While Others Don't?cited by 20
  21. 2025Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectivescited by 15
  22. 2025Predictive Concept Decoders: Training Scalable End-to-End Interpretability Assistantscited by 12
  23. 2025Beyond Chains of Thought: Benchmarking Latent-Space Reasoning Abilities in Large Language Modelscited by 6
  24. 2025From surveillance to signalling: escalation channels as environmental controls for agentic AIcited by 6
  25. 2026Training Agents to Self-Report Misbehaviorcited by 6
  26. 2025Deceptive Automated Interpretability: Language Models Coordinating to Fool Oversight Systemscited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Does Out-of-Sight Equal Out-of-Mind in CoT Monitorability?cited by 0
  2. 2026On the Monitorability of Natural Language Autoencoders: Identification vs Discriminationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026DECOR: Auditing LLM Deception via Information Manipulation Theory
  2. 2026Caught in the Act: Probes Effectively Detect Sabotage and Catch Unverbalized Deception
  3. 2026DeceptGuard :A Constitutional Oversight Framework For Detecting Deception in LLM Agents
  4. 2025Caught in the Act: a mechanistic approach to detecting deception
  5. 2026Lying to Win: Assessing LLM Deception through Human-AI Games and Parallel-World Probing
  6. 2025Beyond Prompt-Induced Lies: Investigating LLM Deception on Benign Prompts
  7. 2026"Did you lie?" Evaluating Lie Detectors across Model Scale and Belief-Verified Model Organisms
  8. 2025Reasoning Isn't Enough: Examining Truth-Bias and Sycophancy in LLMs