arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.07809cs.CRcs.AI
收藏

对人类不可见,由智能体触发:针对移动视觉-语言智能体的隐秘劫持攻击

Invisible to Humans, Triggered by Agents: Stealthy Jailbreak Attacks on Mobile Vision-Language Agents

School of Computer Science, Huazhong University of Science and Technology(华中科技大学计算机科学与技术学院);Dept. of Comp. Sci. and Tech., Institute for AI, Tsinghua-Bosch Joint ML Center, THBI Lab, BNRist Center, Tsinghua University(清华大学计算机科学与技术系、人工智能研究院、清华-博世机器学习联合中心、清华脑与智能实验室、北京国家信息科学与技术研究中心)

Renhua Ding, Xiao Yang, Zhengwei Fang, Jun Luo, Kun He, Jun Zhu

AI总结:

本文提出了一种隐秘的劫持攻击方法,通过智能体交互时的感知注入,避免被人类察觉,同时在移动设备上有效绕过安全过滤器。

摘要:

大型视觉-语言模型(LVLM)赋能自主移动智能体,但其在现实移动部署中的安全性仍待探索。尽管智能体易受视觉提示注入攻击,但隐秘执行此类攻击而不需系统级权限仍具挑战性,因现有方法依赖持久性视觉篡改,易被用户察觉。我们发现人类与智能体交互之间存在一致的不一致:自动化智能体生成几乎零接触触发力信号。基于此见解,我们提出了一种新的攻击范式,即智能体独有感知注入,恶意内容仅在智能体交互期间暴露,而不会被人类用户轻易察觉。为适应移动UI约束和单次交互设置,我们引入HG-IDA*,一种高效的单次优化方法,用于构建绕过LVLM安全过滤器的劫持提示。实验表明,我们的方法可引发未经授权的跨应用操作,在GPT-4o上实现82.5%的规划和75.0%的执行劫持率。我们的发现突显了移动智能体系统中此前未被探索的攻击面,并强调了需要结合交互级信号的防御措施的必要性。

英文摘要:

Large Vision-Language Models (LVLMs) empower autonomous mobile agents, yet their security under realistic mobile deployment constraints remains underexplored. While agents are vulnerable to visual prompt injections, stealthily executing such attacks without requiring system-level privileges remains challenging, as existing methods rely on persistent visual manipulations that are noticeable to users. We uncover a consistent discrepancy between human and agent interactions: automated agents generate near-zero contact touch signals. Building on this insight, we propose a new attack paradigm, agent-only perceptual injection, where malicious content is exposed only during agent interactions, while remaining not readily perceived by human users. To accommodate mobile UI constraints and one-shot interaction settings, we introduce HG-IDA*, an efficient one-shot optimization method for constructing jailbreak prompts that evade LVLM safety filters. Experiments demonstrate that our approach induces unauthorized cross-app actions, achieving 82.5% planning and 75.0% execution hijack rates on GPT-4o. Our findings highlight a previously underexplored attack surface in mobile agent systems and underscore the need for defenses that incorporate interaction-level signals.

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。

↑