arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.19971cs.CR
收藏

先诊断再优化:面向查询高效的越狱攻击的多维评判者反馈

Diagnose Before You Refine: Multi-Dimensional Judge Feedback for Query-Efficient Jailbreak Attacks

Institute of Science Tokyo(东京科学大学);National Yang Ming Chiao Tung University(国立阳明交通大学);Nanyang Technological University(南洋理工大学)

Yixin Tan, Chia-Yi Hsu, Jiayang Liu, Chia-Mu Yu, Rui Wen

AI总结:

该研究针对现有越狱攻击缺乏细粒度诊断信息的问题,提出Jev引导策略老虎机框架JSB,在仅10次查询预算下,于10个开源场景中实现89.8%的平均HarmBench攻击成功率,优于现有方法,还验证了其在不同条件下的有效性。

摘要:

自动化越狱攻击依赖评估反馈来优化对抗性提示,但现有方法常缺乏细粒度诊断信息。我们提出Jev引导策略老虎机(JSB),这是一种基于查询的越狱框架,利用多维概率反馈指导提示优化。JSB结合连续越狱评分与针对性优化、策略切换及最优状态重植,以平衡探索与优化。我们在两个有害请求数据集上,对5个开源和3个闭源大语言模型(LLM)评估JSB。在仅10次目标模型查询的预算下,JSB在10个开源场景中实现平均HarmBench攻击成功率89.8%,始终优于现有自动化越狱攻击。进一步分析表明其在不同查询预算和攻击者模型下均有效,防御实验也揭示了现有过滤机制的局限,并凸显了基于Jev的输出过滤的潜力。

英文摘要:

Automated jailbreak attacks rely on evaluation feedback to optimize adversarial prompts, but existing methods often lack fine-grained diagnostic information. We propose Jev-Guided Strategy Bandit (JSB), a query-based jailbreak framework that leverages multi-dimensional probabilistic feedback to guide prompt optimization. JSB combines continuous jailbreak scoring with targeted refinement, strategy switching, and best-state reseeding to balance exploration and refinement. We evaluate JSB on two harmful request datasets across five open-source and three closed-source LLMs. Under a budget of only 10 target-model queries, JSB achieves an average HarmBench attack success rate of 89.8\% across ten open-source settings, consistently outperforming existing automated jailbreak attacks. Further analyses demonstrate its effectiveness under varying query budgets and attacker models. Defense experiments also reveal limitations in existing filtering mechanisms and highlight the potential of Jev-based output filtering.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑