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 2605.01034cs.CL
收藏

通过组合技能的攻击理论游戏

A Theoretical Game of Attacks via Compositional Skills

University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校);Stony Brook University(石溪大学)

Xinbo Wu, Huan Zhang, Abhishek Umrawal, Lav R. Varshney

AI总结:

本文提出一个理论框架,分析攻击者与防御者之间的博弈,设计最优攻击策略并揭示其与现有对抗提示方法的关系,同时推导出可证明最优的防御策略,并通过实验验证其在不同LLM和基准上的优越性。

摘要:

随着大语言模型能力的提升,其安全部署的担忧加剧。尽管众多对齐策略旨在限制有害行为,但通过精心设计的对抗性提示仍可能被绕过。本文引入了一个理论框架,正式化攻击者与防御者之间的博弈。在此框架中,我们设计了一个理论最佳反应攻击策略,并显示其与许多现有对抗性提示方法密切相关。我们进一步分析了由此产生的博弈,刻画其均衡,并揭示攻击者的内在优势。基于我们的理论分析,我们还推导出一个可证明最优的防御策略。实证上,我们评估了理论最优攻击的实用实例,并在涵盖不同LLM和基准的多样化设置中观察到相对于现有对抗性提示方法的更强性能。

英文摘要:

As large language models grow increasingly capable, concerns about their safe deployment have intensified. While numerous alignment strategies aim to restrict harmful behavior, these defenses can still be circumvented through carefully designed adversarial prompts. In this work, we introduce a theoretical framework that formalizes a game between an attacker and a defender. Within this framework, we design a theoretical best-response attack strategy and show that it is closely related to many existing adversarial prompting methods. We further analyze the resulting game, characterize its equilibria, and reveal inherent advantages for the attacker. Drawing on our theoretical analysis, we also derive a provably optimal defense strategy. Empirically, we evaluate a practical instantiation of the theoretically optimal attack and observe stronger performance relative to existing adversarial prompting approaches in diverse settings encompassing different LLMs and benchmarks.

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。

↑