HarmfulSkillBench: 你的代理如何被有害技能所利用?
HarmfulSkillBench: How Do Harmful Skills Weaponize Your Agents?
AI总结:
本文首次研究了代理生态系统中的有害技能,发现4.93%的技能具有潜在危害,并构建了HarmfulSkillBench基准,评估六个LLM在有害技能下的表现,发现预装技能显著降低拒绝率。
摘要:
大型语言模型(LLMs)已发展为依赖开放技能生态系统的自主代理,如ClawHub和Skills.Rest,托管大量可重用的公开技能。现有安全研究主要关注技能内部的漏洞,如提示注入。然而,关于可能被用于有害行为(如网络攻击、欺诈、隐私侵犯和生成非法内容)的技能,即有害技能,存在关键空白。本文首次对代理生态系统中的有害技能进行了大规模测量研究,涵盖两个主要注册表中的98,440个技能。使用基于我们有害技能分类的LLM驱动评分系统,我们发现4.93%的技能(4,858个)具有危害性,其中ClawHub的有害率高达8.84%,而Skills.Rest为3.49%。我们随后构建了HarmfulSkillBench,首个评估代理在现实场景中对有害技能安全性的基准,包含200个有害技能,涵盖20个类别和四种评估条件。通过评估六个LLM,我们发现通过预装技能呈现有害任务显著降低了所有模型的拒绝率,平均危害分数从无技能时的0.27增加到有技能时的0.47,进一步增加到隐含有害意图时的0.76。我们负责任地向受影响的注册表披露发现,并发布基准以支持未来研究(见https://github.com/TrustAIRLab/HarmfulSkillBench)
英文摘要:
Large language models (LLMs) have evolved into autonomous agents that rely on open skill ecosystems (e.g., ClawHub and Skills.Rest), hosting numerous publicly reusable skills. Existing security research on these ecosystems mainly focuses on vulnerabilities within skills, such as prompt injection. However, there is a critical gap regarding skills that may be misused for harmful actions (e.g., cyber attacks, fraud and scams, privacy violations, and sexual content generation), namely harmful skills. In this paper, we present the first large-scale measurement study of harmful skills in agent ecosystems, covering 98,440 skills across two major registries. Using an LLM-driven scoring system grounded in our harmful skill taxonomy, we find that 4.93% of skills (4,858) are harmful, with ClawHub exhibiting an 8.84% harmful rate compared to 3.49% on Skills.Rest. We then construct HarmfulSkillBench, the first benchmark for evaluating agent safety against harmful skills in realistic agent contexts, comprising 200 harmful skills across 20 categories and four evaluation conditions. By evaluating six LLMs on HarmfulSkillBench, we find that presenting a harmful task through a pre-installed skill substantially lowers refusal rates across all models, with the average harm score rising from 0.27 without the skill to 0.47 with it, and further to 0.76 when the harmful intent is implicit rather than stated as an explicit user request. We responsibly disclose our findings to the affected registries and release our benchmark to support future research (see https://github.com/TrustAIRLab/HarmfulSkillBench).
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。