arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.03255cs.AI
收藏

LPS-Bench: 在长周期规划中评估计算机使用代理的安全意识基准测试

LPS-Bench: Benchmarking Safety Awareness of Computer-Use Agents in Long-Horizon Planning under Benign and Adversarial Scenarios

ShanghaiTech University（上海科技大学）；Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）

Tianyu Chen, Chujia Hu, Dongrui Liu, Xia Hu, Wenjie Wang

备注：
50 pages. Accepted at NeurIPS 2026, Evaluations and Datasets Track. Revised benchmark description, skill-augmented evaluation, evaluator validation, and appendices
阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

LPS-Bench通过评估长周期规划中计算机使用代理的安全意识，揭示现有系统在安全行为维持方面的不足，并提出缓解策略以提升安全性。

摘要：

计算机使用代理（CUAs）通过与真实计算机系统交互可以执行自动化任务，但面临关键的安全风险。模糊的指令可能触发有害行为，而对抗性用户可以操控工具执行以达成恶意目的。现有基准主要集中在短周期或基于图形用户界面的任务上，评估执行时间错误，但忽略了在规划时间预见风险的能力。为填补这一空白，我们提出了LPS-Bench，一个评估基于MCP的CUAs在长周期任务中的规划时间安全意识的基准测试，涵盖65个场景的7个任务领域和9种风险类型。我们引入了一个多代理自动流水线进行可扩展的数据生成，并采用LLM作为裁判的评估协议来通过规划轨迹评估安全意识。实验揭示了现有CUAs在维持安全行为方面存在显著缺陷。我们进一步分析了风险并提出缓解策略以改进基于MCP的CUA系统中的长周期规划安全。我们开源了我们的代码在https://github.com/tychenn/LPS-Bench。

英文摘要：

Computer-use agents (CUAs) execute multi-stage tasks through tools, where an early unsafe decision can propagate to consequential actions. Evaluating only final outcomes can miss such decisions, while constructing executable environments for new tasks can make benchmark expansion costly. We present LPS-Bench, a benchmark of long-horizon planning safety in MCP-style tool workflows under benign requests and adversarial steering. A template-guided multi-agent pipeline generates user instructions, simulated toolkits, and case-specific safety criteria, followed by human review. This design supports scalable case expansion without building a separate application environment for every test case. LPS-Bench comprises 570 cases derived from 65 scenarios across 7 task domains and 9 planning-risk types, with representative cases additionally adapted to reusable skills. An LLM-based evaluator applies case-specific criteria to complete interaction records, examining tool choices, arguments, and responses to environmental feedback throughout execution. Evaluations of 13 LLM agents reveal persistent failures in both benign and adversarial settings. Prompt-based interventions yield model-dependent gains, but substantial safety failures remain.

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。

↑