arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.03081cs.CL
收藏

TAO-Attack:迈向大型语言模型的高级优化基 Jailbreak 攻击

TAO-Attack: Toward Advanced Optimization-Based Jailbreak Attacks for Large Language Models

Dalian University of Technology(大连理工大学)

Zhi Xu, Jiaqi Li, Xiaotong Zhang, Hong Yu, Han Liu

AI总结:

TAO-Attack 提出了一种基于优化的 Jailbreak 攻击方法,通过双阶段损失函数和方向优先 token 优化策略,有效提升攻击成功率。

摘要:

大型语言模型(LLMs)在各种应用中取得了显著成功,但仍然容易受到 Jailbreak 攻击的威胁,其中攻击者会构造提示以绕过安全对齐并引发不安全响应。在现有方法中,基于优化的攻击已显示出强大的有效性,但当前方法常常面临频繁拒绝、伪有害输出和低效的 token 级更新。在本工作中,我们提出了 TAO-Attack,一种新的基于优化的 Jailbreak 方法。TAO-Attack 采用双阶段损失函数:第一阶段抑制拒绝以确保模型继续产生有害前缀,而第二阶段惩罚伪有害输出并鼓励模型朝更有害的完成发展。此外,我们设计了一种方向优先 token 优化(DPTO)策略,通过在考虑更新幅度之前对候选者与梯度方向对齐来提高效率。在多个 LLMs 上的广泛实验表明,TAO-Attack 一致优于最先进的方法,实现了更高的攻击成功率,甚至在某些场景中达到 100%。

英文摘要:

Large language models (LLMs) have achieved remarkable success across diverse applications but remain vulnerable to jailbreak attacks, where attackers craft prompts that bypass safety alignment and elicit unsafe responses. Among existing approaches, optimization-based attacks have shown strong effectiveness, yet current methods often suffer from frequent refusals, pseudo-harmful outputs, and inefficient token-level updates. In this work, we propose TAO-Attack, a new optimization-based jailbreak method. TAO-Attack employs a two-stage loss function: the first stage suppresses refusals to ensure the model continues harmful prefixes, while the second stage penalizes pseudo-harmful outputs and encourages the model toward more harmful completions. In addition, we design a direction-priority token optimization (DPTO) strategy that improves efficiency by aligning candidates with the gradient direction before considering update magnitude. Extensive experiments on multiple LLMs demonstrate that TAO-Attack consistently outperforms state-of-the-art methods, achieving higher attack success rates and even reaching 100\% in certain scenarios.

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。

↑