RL-Obfuscation: Can Language Models Learn to Evade Latent-Space Monitors? (2506.14261v2)

Abstract: Latent-space monitors aim to detect undesirable behaviours in LLMs by leveraging internal model representations rather than relying solely on black-box outputs. These methods have shown promise in identifying behaviours such as deception and unsafe completions, but a critical open question remains: can LLMs learn to evade such monitors? To study this, we introduce RL-Obfuscation, in which LLMs are finetuned via reinforcement learning to bypass latent-space monitors while maintaining coherent generations. We apply RL-Obfuscation to LLMs ranging from 7B to 14B parameters and evaluate evasion success against a suite of monitors. We find that token-level latent-space monitors are highly vulnerable to this attack. More holistic monitors, such as max-pooling or attention-based probes, remain robust. Moreover, we show that adversarial policies trained to evade a single static monitor generalise to unseen monitors of the same type. Finally, we study how the policy learned by RL bypasses these monitors and find that the model can also learn to repurpose tokens to mean something different internally.