PulseAugur
实时 13:18:06
English(EN) When Prompts Control Robots: Prompt Injection Attacks in Multi-Agent Robotic Systems

物理提示词注入攻击危及视觉语言模型控制的机器人

研究人员调查了视觉语言模型(VLM)控制的机器人中的提示词注入攻击。第一项研究系统地检查了使用机器人视野中的对抗性文本进行的物理提示词注入,对攻击进行了分类,并在三个前沿 VLM(GPT-4o、Gemini 2.5 Flash、Qwen3-VL-32B)上进行了测试。这些攻击的成功率在 5% 到 29.4% 之间,其中冒充权威和否定攻击被证明具有跨模型可转移性。第二项研究侧重于多智能体机器人系统,证明提示词注入会导致对抗性行为并在智能体之间传播,影响任务完成和安全性。两项研究都强调了 VLM 控制的机器人对此类攻击的脆弱性,并探讨了潜在的缓解策略。 AI

影响 凸显了 AI 控制的机器人技术中重大的安全漏洞,亟需开发针对对抗性操纵的强大防御措施。

排序理由 该集群包含两篇发表在 arXiv 上的学术论文,详细介绍了对视觉语言模型控制的机器人进行提示词注入攻击的研究。

在 arXiv cs.MA (Multiagent) 阅读 →

AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →

物理提示词注入攻击危及视觉语言模型控制的机器人

报道来源 [4]

  1. arXiv cs.AI TIER_1 English(EN) · S. M . Bhagya P. Samarakoon, M. A. Viraj J. Muthugala, W. K. R. Sachinthana, Mohan Rajesh Elara ·

    用一张纸劫持机器人:VLM控制机器人物理提示注入的系统研究

    arXiv:2608.05715v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) are increasingly deployed as planners in robotic systems, where they translate natural-language commands into executable actions grounded in visual scene understanding. This tight coupling between per…

  2. arXiv cs.AI TIER_1 English(EN) · Neha Nagaraja, Amisha Bagari, Hayretdin Bahsi ·

    当提示词控制机器人:多智能体机器人系统中的提示注入攻击

    arXiv:2608.00747v2 Announce Type: replace-cross Abstract: Large language models are increasingly integrated into autonomous robotic systems for task planning and control, but this integration exposes them to prompt injection attacks that can lead to unsafe decisions and physical …

  3. arXiv cs.MA (Multiagent) TIER_1 English(EN) · Hayretdin Bahsi ·

    当提示词控制机器人:多智能体机器人系统中的提示注入攻击

    Large language models are increasingly integrated into autonomous robotic systems for task planning and control, but this integration exposes them to prompt injection attacks that can lead to unsafe decisions and physical harm. Multi-agent settings increase the risks through cros…

  4. arXiv cs.MA (Multiagent) TIER_1 English(EN) · Hayretdin Bahsi ·

    当提示词控制机器人:多智能体机器人系统中的提示注入攻击

    Large language models are increasingly integrated into autonomous robotic systems for task planning and control, but this integration exposes them to prompt injection attacks that can lead to unsafe decisions and physical harm. Multi-agent settings increase the risks through cros…