A False Average: Pooled CoT-Monitor Accuracy Conceals a Reasoning-Dependent Fragility
cs.CR, cs.AI, cs.CL, cs.LG
Submitted: 2026-08-01
Updated: 2026-09-11
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- How does information access affect LLM monitors' ability to detect sabotage?
- Obfuscated Activations Bypass LLM Latent-Space Defenses
- Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
- Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety
- Terminal Wrench: A Dataset of 331 Reward-Hackable Environments and 3,632 Exploit Trajectories
- Building Production-Ready Probes For Gemini
- Measuring Faithfulness in Chain-of-Thought Reasoning
- Natural Emergent Misalignment from Reward Hacking in Production RL
- Where Reliability Lives in Vision-Language Models: A Mechanistic Study of Attention, Hidden States, and Causal Circuits
- Reasoning Models Don't Always Say What They Think
- Visuals Lie, Consistency Speaks: Disentangling Spatial Attention from Reliability in Vision-Language Models
- Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models
- When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors
- Benchmarking Deception Probes via Black-to-White Performance Boosts
- But what is your honest answer? Aiding LLM-judges with honest alternatives using steering vectors
- COMPASS: Context-Modulated PID Attention Steering System for Hallucination Mitigation
- Detecting Strategic Deception Using Linear Probes
- Linear Predictability of Attention Heads in Large Language Models
- RL-Obfuscation: Can Language Models Learn to Evade Latent-Space Monitors?
- The Hypocrisy Gap: Quantifying Divergence Between Internal Belief and Chain-of-Thought Explanation via Sparse Autoencoders
Related papers
- SoK: AI-Augmented Binary Reversing
- Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
- Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
- Efficient Fuzzy PSI under One-Sided Assumptions
- Sealing the Audit-Runtime Gap for LLM Skills
- Token Composition: A Graph Based on EVM Logs