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 2510.02554cs.CRcs.AI
收藏

ToolTweak:对基于大语言模型的智能体中工具选择的攻击

ToolTweak: An Attack on Tool Selection in LLM-based Agents

University of Oxford(牛津大学);Microsoft(微软公司)

Jonathan Sneh, Ruomei Yan, Jialin Yu, Philip Torr, Yarin Gal, Sunando Sengupta, Eric Sommerlade, Alasdair Paren, Adel Bibi

AI总结:

本文针对基于LLM的智能体工具选择过程的漏洞,提出轻量级自动攻击ToolTweak,可大幅提升特定工具选择率且跨模型迁移性强,同时评估了两种有效防御手段,为维护工具生态安全提供参考。

摘要:

随着大语言模型(LLM)越来越多地为与外部工具交互的智能体提供支持,工具使用已成为扩展其能力的核心机制。这类智能体通常从不断扩充的工具数据库或市场中选择工具来完成用户任务,这使得工具提供商与开发者之间为获得曝光度和使用率形成了隐性竞争。本文指出,这一选择过程存在关键漏洞:攻击者可通过迭代操纵工具名称与描述,系统性地引导智能体偏向选择特定工具,从而在功能相当的同类工具中获取不公平优势。我们提出了ToolTweak这一轻量级自动攻击方法,它能将工具的基线选择率从约20%提升至最高81%,且在开源与闭源模型间具备强迁移性。除单个工具外,我们还证明此类攻击会导致工具使用分布发生偏移,给新兴工具生态系统的公平性、竞争性与安全性带来风险。为缓解这些风险,我们评估了两种防御手段:释义改写与困惑度过滤,它们可降低选择偏差,使智能体更均衡地选择功能相近的工具。所有代码将在论文录用后开源。

英文摘要:

As LLMs increasingly power agents that interact with external tools, tool use has become an essential mechanism for extending their capabilities. These agents typically select tools from growing databases or marketplaces to solve user tasks, creating implicit competition among tool providers and developers for visibility and usage. In this paper, we show that this selection process harbors a critical vulnerability: by iteratively manipulating tool names and descriptions, adversaries can systematically bias agents toward selecting specific tools, gaining unfair advantage over equally capable alternatives. We present ToolTweak, a lightweight automatic attack that increases selection rates from a baseline of around 20% to as high as 81%, with strong transferability between open-source and closed-source models. Beyond individual tools, we show that such attacks cause distributional shifts in tool usage, revealing risks to fairness, competition, and security in emerging tool ecosystems. To mitigate these risks, we evaluate two defenses: paraphrasing and perplexity filtering, which reduce bias and lead agents to select functionally similar tools more equally. All code will be open-sourced upon acceptance.

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。

↑