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 2601.03288cs.CRcs.CL
收藏

你的 jailbreak 真的可信吗?基于锚定参考的细粒度 jailbreak 评估

How Real is Your Jailbreak? Fine-grained Jailbreak Evaluation with Anchored Reference

Key Laboratory of Trustworthy Distributed Computing(可信分布式计算重点实验室);Service Beijing University of Posts(北京邮电大学);Artificial Intelligence Research Institute China Academy of Information(人工智能研究所信息与通信技术 academy)

Songyang Liu, Chaozhuo Li, Rui Pu, Litian Zhang, Chenxu Wang, Zejian Chen, Yuting Zhang, Yiming Hei

AI总结:

本文提出 FJAR 框架,通过细粒度分类和锚定参考方法,评估 jailbreak 攻击的真实性和失败原因,提升攻击策略改进的指导性。

摘要:

jailbreak 攻击对大型语言模型(LLMs)的安全性构成了重大挑战,但当前的自动化评估方法大多依赖于粗略分类,主要关注有害性,导致攻击成功率的严重高估。为了解决这个问题,我们提出了 FJAR,一种基于锚定参考的细粒度 jailbreak 评估框架。我们首先将 jailbreak 响应分为五个细粒度类别:拒绝、不相关、无帮助、错误和成功,基于响应是否回应恶意查询的意图程度。这种分类构成了 FJAR 的基础。然后,我们介绍了一种新的无害树分解方法,通过分解原始查询来构建高质量的锚定参考。这些参考指导评估者判断响应是否真正满足原始查询。广泛实验表明,FJAR 与人类判断的匹配度最高,并能有效识别 jailbreak 失败的根本原因,为改进攻击策略提供可行指导。

英文摘要:

Jailbreak attacks present a significant challenge to the safety of Large Language Models (LLMs), yet current automated evaluation methods largely rely on coarse classifications that focus mainly on harmfulness, leading to substantial overestimation of attack success. To address this problem, we propose FJAR, a fine-grained jailbreak evaluation framework with anchored references. We first categorized jailbreak responses into five fine-grained categories: Rejective, Irrelevant, Unhelpful, Incorrect, and Successful, based on the degree to which the response addresses the malicious intent of the query. This categorization serves as the basis for FJAR. Then, we introduce a novel harmless tree decomposition approach to construct high-quality anchored references by breaking down the original queries. These references guide the evaluator in determining whether the response genuinely fulfills the original query. Extensive experiments demonstrate that FJAR achieves the highest alignment with human judgment and effectively identifies the root causes of jailbreak failures, providing actionable guidance for improving attack strategies.

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。

↑