From Hidden States to Confessions: Auditing Deception in Reasoning Models
cs.CL, cs.AI
Submitted: 2026-06-16
Updated: 2026-11-07
Code: https://github.com/March-7/StateWitness
Terminology
Sources
- Chain-of-Thought Reasoning In The Wild Is Not Always Faithful
- The Internal State of an LLM Knows When It's Lying
- Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
- Towards evaluations-based safety cases for AI scheming
- Chain-of-Thought Unfaithfulness as Disguised Accuracy
- Sabotage Evaluations for Frontier Models
- Discovering Latent Knowledge in Language Models Without Supervision
- Scheming AIs: Will AIs fake alignment during training in order to get power?
- AI Deception: Risks, Dynamics, and Controls
- Reasoning Models Don't Always Say What They Think
- ELEPHANT: Measuring and understanding social sycophancy in LLMs
- BeHonest: Benchmarking Honesty in Large Language Models
- Are DeepSeek R1 And Other Reasoning Models More Faithful?
- Eliciting Secret Knowledge from Language Models
- SycEval: Evaluating LLM Sycophancy
- From surveillance to signalling: escalation channels as environmental controls for agentic AI
- Alignment faking in large language models
- AI Control: Improving Safety Despite Intentional Subversion
- Beyond Chains of Thought: Benchmarking Latent-Space Reasoning Abilities in Large Language Models
- An Assessment of Model-On-Model Deception
Related papers
- ALARM: Audio-Language Alignment for Reasoning Models
- ThinkSum: Probabilistic reasoning over sets using large language models
- Coherence boosting: When your pretrained language model is not paying enough attention
- From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for Role-Playing Agents
- Improving General Role-Playing Agents via Psychology-Grounded Reasoning and Role-Aware Policy Optimization
- The Art of Asking: Multilingual Prompt Optimization for Synthetic Data