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 2603.18762cs.CRcs.AI
收藏

ClawTrap:一种基于中间人攻击的红队框架,用于现实世界OpenClaw安全评估

ClawTrap: A MITM-Based Red-Teaming Framework for Real-World OpenClaw Security Evaluation

Haochen Zhao, Shaoyang Cui

AI总结:

本文提出ClawTrap,一种基于中间人攻击的红队框架,用于评估现实世界OpenClaw的安全性。该框架支持多种定制化攻击形式,提供可重复的管道流程,用于规则驱动的拦截、转换和审计。

摘要:

自主网络代理如OpenClaw正迅速进入高影响的现实世界工作流程,但其在真实网络威胁下的安全性仍缺乏充分评估。现有基准主要关注静态沙箱设置和内容级提示攻击,这在网络层安全测试方面存在实际缺口。本文提出了ClawTrap,一种基于中间人攻击的红队框架,用于现实世界OpenClaw安全评估。ClawTrap支持多种多样且可定制的攻击形式,包括静态HTML替换、iFrame弹出注入和动态内容修改,并提供可重复的管道流程用于规则驱动的拦截、转换和审计。该设计为未来研究构建更丰富、可定制的中间人攻击奠定了基础,并能系统地对代理框架和模型后端进行安全测试。我们的实证研究表明清晰的模型分层:较弱的模型更可能信任篡改的观察并产生不安全的输出,而更强的模型则表现出更好的异常归因和更安全的回退策略。这些发现表明,可靠的OpenClaw安全评估应明确纳入动态现实世界中间人攻击条件,而不是仅依赖静态沙箱协议。

英文摘要:

Autonomous web agents such as \textbf{OpenClaw} are rapidly moving into high-impact real-world workflows, but their security robustness under live network threats remains insufficiently evaluated. Existing benchmarks mainly focus on static sandbox settings and content-level prompt attacks, which leaves a practical gap for network-layer security testing. In this paper, we present \textbf{ClawTrap}, a \textbf{MITM-based red-teaming framework for real-world OpenClaw security evaluation}. ClawTrap supports diverse and customizable attack forms, including \textit{Static HTML Replacement}, \textit{Iframe Popup Injection}, and \textit{Dynamic Content Modification}, and provides a reproducible pipeline for rule-driven interception, transformation, and auditing. This design lays the foundation for future research to construct richer, customizable MITM attacks and to perform systematic security testing across agent frameworks and model backbones. Our empirical study shows clear model stratification: weaker models are more likely to trust tampered observations and produce unsafe outputs, while stronger models demonstrate better anomaly attribution and safer fallback strategies. These findings indicate that reliable OpenClaw security evaluation should explicitly incorporate dynamic real-world MITM conditions rather than relying only on static sandbox protocols.

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。

↑