DELMAN: Dynamic Defense Against Large Language Model Jailbreaking with Model Editing
This addresses the critical safety issue of jailbreak attacks for deployed LLMs, though it appears incremental as an improvement over existing model editing techniques.
The paper tackles the problem of jailbreak attacks on large language models by proposing DELMAN, a dynamic defense method that directly edits a minimal set of model parameters to neutralize harmful behaviors while preserving utility. Experimental results show it outperforms baseline methods in mitigating attacks and adapting to new instances.
Large Language Models (LLMs) are widely applied in decision making, but their deployment is threatened by jailbreak attacks, where adversarial users manipulate model behavior to bypass safety measures. Existing defense mechanisms, such as safety fine-tuning and model editing, either require extensive parameter modifications or lack precision, leading to performance degradation on general tasks, which is unsuitable to post-deployment safety alignment. To address these challenges, we propose DELMAN (Dynamic Editing for LLMs JAilbreak DefeNse), a novel approach leveraging direct model editing for precise, dynamic protection against jailbreak attacks. DELMAN directly updates a minimal set of relevant parameters to neutralize harmful behaviors while preserving the model's utility. To avoid triggering a safe response in benign context, we incorporate KL-divergence regularization to ensure the updated model remains consistent with the original model when processing benign queries. Experimental results demonstrate that DELMAN outperforms baseline methods in mitigating jailbreak attacks while preserving the model's utility, and adapts seamlessly to new attack instances, providing a practical and efficient solution for post-deployment model protection.