Paper Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitors · 2025 · arXiv:2512.11949 Read the paper → deceptionmech-interpmonitoring
