Search papers, labs, and topics across Lattice.
This paper introduces a self-evolving defense mechanism against jailbreak attacks on large language models (LLMs) that leverages a persistent, cross-interaction rule memory to adaptively respond to emerging threats. By abstracting successful attack patterns into method-level rules, the framework generalizes defenses across entire attack families without requiring parameter updates. Experimental results demonstrate significant reductions in attack success rates across various models and jailbreak families, while maintaining benign utility and robustness against adaptive attacks.
Attack success rates drop significantly as this framework evolves its defenses in real-time, adapting to new jailbreak strategies without retraining.
Large language models (LLMs) remain vulnerable to jailbreak attacks that exploit techniques such as role-playing, obfuscation, code transformation, and multi-step indirection to elicit harmful outputs. As jailbreak strategies keep emerging, defenses have proliferated in an ongoing cat-and-mouse game, yet most remain static: their safety behavior is fixed at deployment, so they cannot accumulate defensive experience or adapt to unseen strategies. We propose a self-evolving test-time defense built around a persistent, cross-interaction rule memory: when an attack succeeds, the framework abstracts that failure into a method-level rule capturing the structural attack wrapper rather than the harmful topic, and reuses it against future inputs. Because rules are method-level, one induced rule generalizes across an entire attack family, and the label space expands as novel wrappers appear. The mechanism operates entirely through external memory and prompting, with no parameter updates, and applies to both open-weight and black-box API models. We realize it as four cooperating modules, but the contribution is the memory-based adaptation mechanism, not the module decomposition. Across four black-box jailbreak families and multiple models, our method substantially reduces attack success rates while preserving benign utility, remains robust under an adaptive composite-wrapper attack, and does not increase over-refusal as the memory grows.
