arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2310.04451cs.CLcs.AI
收藏

AutoDAN:在对齐的大型语言模型上生成隐蔽的越狱提示

AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models

University of Wisconsin–Madison(威斯康星大学麦迪逊分校);USC(南加州大学);University of California, Davis(加利福尼亚大学戴维斯分校)

Xiaogeng Liu, Nan Xu, Muhao Chen, Chaowei Xiao

AI总结:

本文提出AutoDAN,一种利用分层遗传算法自动生成隐蔽越狱提示的新型攻击方法,解决了现有技术可扩展性与隐蔽性问题,在跨模型和跨样本攻击中表现卓越并能有效绕过困惑度防御。

摘要:

通过对齐的大型语言模型(LLMs)是通过与人类反馈的广泛对齐而创建的强大语言理解和决策工具。然而,这些大型模型仍然容易受到越狱攻击,攻击者操纵提示以引发对齐LLMs不应给出的恶意输出。研究越狱提示可以引导我们深入探讨LLMs的局限性,并进一步指导我们保护它们。不幸的是,现有的越狱技术要么存在(1)可扩展性问题,攻击严重依赖手工制作提示,要么存在(2)隐蔽性问题,因为攻击依赖基于token的算法生成提示,这些提示通常在语义上无意义,使其容易通过基本的困惑度测试被检测到。鉴于这些挑战,我们打算回答这个问题:我们能否开发一种方法,自动生成隐蔽的越狱提示?在本文中,我们介绍了AutoDAN,一种针对对齐LLMs的新型越狱攻击。AutoDAN可以通过精心设计的分层遗传算法自动生成隐蔽的越狱提示。广泛的评估表明,AutoDAN不仅自动化了该过程并保持了语义的有意义性,而且与基线相比,在跨模型可迁移性和跨样本通用性方面表现出卓越的攻击强度。此外,我们还比较了AutoDAN与基于困惑度的防御方法,并表明AutoDAN可以有效地绕过它们。

英文摘要:

The aligned Large Language Models (LLMs) are powerful language understanding and decision-making tools that are created through extensive alignment with human feedback. However, these large models remain susceptible to jailbreak attacks, where adversaries manipulate prompts to elicit malicious outputs that should not be given by aligned LLMs. Investigating jailbreak prompts can lead us to delve into the limitations of LLMs and further guide us to secure them. Unfortunately, existing jailbreak techniques suffer from either (1) scalability issues, where attacks heavily rely on manual crafting of prompts, or (2) stealthiness problems, as attacks depend on token-based algorithms to generate prompts that are often semantically meaningless, making them susceptible to detection through basic perplexity testing. In light of these challenges, we intend to answer this question: Can we develop an approach that can automatically generate stealthy jailbreak prompts? In this paper, we introduce AutoDAN, a novel jailbreak attack against aligned LLMs. AutoDAN can automatically generate stealthy jailbreak prompts by the carefully designed hierarchical genetic algorithm. Extensive evaluations demonstrate that AutoDAN not only automates the process while preserving semantic meaningfulness, but also demonstrates superior attack strength in cross-model transferability, and cross-sample universality compared with the baseline. Moreover, we also compare AutoDAN with perplexity-based defense methods and show that AutoDAN can bypass them effectively.

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。

↑