What Intermediate Layers Know: Detecting Jailbreaks from Entropy Dynamics
For LLM safety researchers, this work reveals that jailbreak-relevant structure is most pronounced in mid-network representations, offering a new detection signal that does not require model modification.
The paper investigates how jailbreak attacks are encoded in LLM internal representations by analyzing token-level predictive entropy trajectories across layers. It finds that entropy dynamics, particularly monotonic rank-based trend scores in intermediate layers, provide architecture-consistent separation between jailbreak and benign prompts without additional training.
Jailbreak attacks reveal a persistent weakness in aligned Large Language Models: carefully crafted prompts can elicit policy-violating responses despite safety training. While most defenses operate at the prompt or output level, it remains unclear how harmful intent is encoded within the model's internal representations. We investigate this question by analyzing token-level predictive entropy trajectories across layers of a frozen LLM using the logit lens. We find that static aggregate statistics of prompt-level entropy (e.g., mean, variance) carry little discriminative signal, whereas features capturing how entropy evolves across token positions, such as monotonic rank-based trend scores, are substantially more informative. Importantly, this signal is not uniform across model depth: it is concentrated in intermediate layers and degrades at the final layer, indicating that jailbreak-relevant structure is most pronounced in mid-network representations rather than at the output head. Across multiple models (Llama, Qwen, Gemma) and adversarial benchmarks, these entropy dynamics provide architecture-consistent separation without additional training. Together, our findings show that jailbreak behavior is reflected in structured intermediate uncertainty dynamics, clarifying both which entropy-derived features encode harmful intent and where in the network that signal is most pronounced.