Evaluating Privilege Usage of Agents with Real-World Tools
- ID: 06322165
- 原文链接: https://arxiv.org/abs/2603.28166
- PDF: https://arxiv.org/pdf/2603.28166v2
- 作者: Quan Zhang, Lianhang Fu, Lvsi Lian, Gwihwan Go, Yujue Wang, Chijin Zhou, Yu Jiang, Geguang Pu
- 日期: 2026-03-30
- 更新: 2026-04-20
- 分类: agents
- 来源类型: paper
- 标签: agent-evaluation, security-benchmark, real-world-tools, privilege-usage, cs.cr
- 质量评分: 4/5
- 抓取时间: 2026-07-25T04:19:42+00:00
中文导读
GrantBox 论文指出现有 Agent 安全 benchmark 常依赖预编码工具和受限交互，难以代表真实工具环境摘要提出一个面向真实世界工具的安全评估沙箱，用来观察 Agent 如何使用继承而来的工具权限以及由此导致的信息泄露或基础设施损害风险
为什么值得关注
评估 Agent 权限风险需要真实工具沙箱；玩具工具环境很容易低估泄露和破坏面
这篇论文属于 AAIF 的 Agent 安全/评估线索：它不是泛泛讨论“让模型更安全”，而是把风险落到工具调用、权限继承、授权边界或集合级效用等可被工程系统观测与约束的对象上。对于正在构建工具型 Agent、连接器、沙箱评测或推荐型 Agent 的团队，它提供了一个更具体的检查维度：模型输出之外，系统还需要回答“Agent 被允许做什么、实际做了什么、这些动作对用户账户/环境/推荐集合造成什么后果”。
关键信息
- 论文标题：Evaluating Privilege Usage of Agents with Real-World Tools
- 作者：Quan Zhang, Lianhang Fu, Lvsi Lian, Gwihwan Go, Yujue Wang, Chijin Zhou, Yu Jiang, Geguang Pu
- arXiv：https://arxiv.org/abs/2603.28166
- 发布时间：2026-03-30
- arXiv 分类：cs.CR, cs.AI
- 关联标签：agent-evaluation, security-benchmark, real-world-tools, privilege-usage, cs.cr
English Abstract
Equipping LLM agents with real-world tools can substantially improve productivity. However, granting agents autonomy over tool use also transfers the associated privileges to both the agent and the underlying LLM. Improper privilege usage may lead to serious consequences, including information leakage and infrastructure damage. While several benchmarks have been built to study agents' security, they often rely on pre-coded tools and restricted interaction patterns. Such crafted environments differ substantially from the real-world, making it hard to assess agents' security capabilities in critical privilege control and usage. Therefore, we propose GrantBox, a security evaluation sandbox for analyzing agent privilege usage. GrantBox automatically integrates real-world tools and allows LLM agents to invoke genuine privileges, enabling the evaluation of privilege usage under prompt injection attacks. Our results indicate that while LLMs exhibit basic security awareness and can block some direct attacks, they remain vulnerable to more sophisticated attacks, resulting in an average attack success rate of 84.80% in carefully crafted scenarios.
English Summary
Equipping LLM agents with real-world tools can substantially improve productivity. However, granting agents autonomy over tool use also transfers the associated privileges to both the agent and the underlying LLM. Improper privilege usage may lead to serious consequences, including information leakage and infrastructure damage. While several benchmarks have been built to study agents' security, they often rely on pre-coded tools and restricted interaction patterns. Such crafted environments differ substantially from the real-world, making it hard to assess agents' security capabilities in critical privilege control and usage. Therefore, we propose GrantBox, a security evaluation sandbox for analyzing agent privilege usage. GrantBox automatically integrates real-world tools and allows LLM agents to invoke genuine privileges, enabling the evaluation of privilege usage under prompt injection attacks. Our results indicate that while LLMs exhibit basic security awareness and can block some direct attacks, they remain vulnerable to more sophisticated attacks, resulting in an average attack success rate of 84.80% in carefully crafted scenarios.
Obsidian Notes
- 内容由
opencli arxiv paper拉取 arXiv 元数据与摘要生成。 - 中文导读与价值判断均锚定在条目已有摘要、论文摘要、作者、日期与分类信息上；未补充论文摘要之外的实验细节。