DualMirage：利用轮廓与对抗错觉的 CAPTCHA 猎捕隐蔽多模态大语言模型智能体
DualMirage: Hunting Stealthy Multimodal LLM Agents via CAPTCHAs with Contour and Adversarial Illusions
利用人类与机器感知差异设计 CAPTCHA，通过轮廓错觉和对抗扰动探测 MLLM 智能体，为视觉对抗攻击与多模态安全评估提供新思路。
原文依据 DualMirage employs a dual-pronged strategy: (1) Contour Illusions, which utilize cognitive principles to generate illusory contours that humans perceive effortlessly yet pose interpretation challenges for MLLMs; and (2) Adversarial Illusions, which embed human-imperceptible perturbations optimized to mislead the visual encoders of target MLLMs
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论多模态大语言模型（MLLMs）的快速发展催生了能够执行复杂类人任务的先进自主智能体。然而，这也引入了显著的安全风险，尤其是那些通过模仿人类行为来规避传统检测机制的隐蔽 MLLM 智能体。本文提出 DualMirage，一种新颖的 CAPTCHA 框架，通过利用人类与机器感知之间的根本差异来主动对抗和识别隐蔽智能体。DualMirage 采用双管齐下的策略：（1）轮廓错觉，利用认知原理生成人类可轻松感知但对 MLLM 构成解读挑战的错觉轮廓；（2）对抗错觉，嵌入人类不可察觉的扰动，这些扰动经过优化以误导目标 MLLM 的视觉编码器，从而引发特征性、可识别的模型响应。在五个最优 MLLM 上的评估表明，DualMirage 平均达到 95.8% 的人类成功率，同时阻止 MLLM 智能体（智能体阻止率高达 100%），优于现有 CAPTCHA。此外，DualMirage 诱导模型主动暴露身份，实现 58.8% 的白盒和 21.9% 的黑盒攻击成功率，证明其对隐蔽多模态智能体有效。
英文原文摘要 1364 字符
The rapid advancement of Multimodal Large Language Models (MLLMs) has given rise to sophisticated autonomous agents capable of performing complex, human-like tasks across the web. However, this also introduces significant security risks, particularly from stealthy MLLM agents that can evade conventional detection mechanisms by mimicking human behavior. In this paper, we propose DualMirage, a novel CAPTCHA framework that proactively counters and identifies stealthy agents by exploiting fundamental disparities between human and machine perception. DualMirage employs a dual-pronged strategy: (1) Contour Illusions, which utilize cognitive principles to generate illusory contours that humans perceive effortlessly yet pose interpretation challenges for MLLMs; and (2) Adversarial Illusions, which embed human-imperceptible perturbations optimized to mislead the visual encoders of target MLLMs and thereby elicit characteristic, identifiable model responses. Evaluations on five state-of-the-art MLLMs demonstrate that DualMirage achieves an average 95.8% human success rate while blocking MLLM agents (up to 100% agent blocking rate), outperforming existing CAPTCHAs. Furthermore, DualMirage induces models to expose identities actively, achieving 58.8% white-box and 21.9% black-box attack success rates, proving effective against stealthy multimodal agents.