LLMs Learn to Evade Latent Monitors from Prior Feedback Alone
cs.LG, cs.AI
Submitted: 2026-09-29
Updated: 2026-09-29
Sources
- Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
- Training Verifiers to Solve Math Word Problems
- Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models
- Gemma 3 Technical Report
- The Llama 3 Herd of Models
- RL-Obfuscation: Can Language Models Learn to Evade Latent-Space Monitors?
- Risks from Learned Optimization in Advanced Machine Learning Systems
- Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
- Building Production-Ready Probes For Gemini
- Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitors
- Steering Awareness: Detecting Activation Steering from Within
- Resummation of the C-Parameter Sudakov Shoulder Using Effective Field Theory
- Emotion Concepts and their Function in a Large Language Model
- Prefill Awareness in Large Language Models
- Qwen3 Technical Report
- Representation Engineering: A Top-Down Approach to AI Transparency
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks