arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.00565cs.CVcs.AIcs.CR
收藏

MIDAS: 多图像分散与语义重建用于对抗多模态大语言模型

MIDAS: Multi-Image Dispersion and Semantic Reconstruction for Jailbreaking MLLMs

Beijing University of Posts and Telecommunications(北京邮电大学);Nanyang Technological University(南洋理工大学);Guilin University of Electronic Technology(桂林电子科技大学)

Yilian Liu, Xiaojun Jia, Guoshun Nan, Jiuyang Lyu, Zhican Chen, Tao Guan, Shuyuan Luo, Zhongyi Zhai, Yang Liu

AI总结:

MIDAS通过多图像分散与语义重建技术,提升对抗多模态大语言模型的劫持性能,达到81.46%的平均攻击成功率。

摘要:

多模态大语言模型(MLLMs)已取得显著性能,但仍然容易受到劫持攻击,这些攻击可能导致有害内容并破坏其安全部署。先前研究显示,引入额外的推理步骤,这些步骤会破坏安全注意力,会使MLLMs更容易被误导生成恶意内容。然而,这些方法依赖于单图像遮蔽或孤立的视觉提示,仅略微扩展推理路径,因此效果有限,特别是针对高度对齐的商业闭源模型。为了解决这个问题,本文提出了多图像分散与语义重建(MIDAS),一种多模态劫持框架,将有害语义分解为风险承担的子单元,将其分散到多个视觉提示中,并利用跨图像推理逐步重建恶意意图,从而绕过现有的安全机制。所提出的MIDAS强制更长且更结构化的多图像链式推理,显著增加模型对视觉提示的依赖性,同时延迟恶意语义的暴露,并显著减少模型的安全注意力,从而提高对先进MLLMs的劫持性能。在不同数据集和MLLMs上的广泛实验表明,所提出的MIDAS在MLLMs的劫持攻击中优于最先进的劫持攻击,并在4个闭源MLLMs上实现了平均攻击成功率81.46%。我们的代码可在[此链接](https://github.com/Winnie-Lian/MIDAS)获取。

英文摘要:

Multimodal Large Language Models (MLLMs) have achieved remarkable performance but remain vulnerable to jailbreak attacks that can induce harmful content and undermine their secure deployment. Previous studies have shown that introducing additional inference steps, which disrupt security attention, can make MLLMs more susceptible to being misled into generating malicious content. However, these methods rely on single-image masking or isolated visual cues, which only modestly extend reasoning paths and thus achieve limited effectiveness, particularly against strongly aligned commercial closed-source models. To address this problem, in this paper, we propose Multi-Image Dispersion and Semantic Reconstruction (MIDAS), a multimodal jailbreak framework that decomposes harmful semantics into risk-bearing subunits, disperses them across multiple visual clues, and leverages cross-image reasoning to gradually reconstruct the malicious intent, thereby bypassing existing safety mechanisms. The proposed MIDAS enforces longer and more structured multi-image chained reasoning, substantially increases the model's reliance on visual cues while delaying the exposure of malicious semantics and significantly reducing the model's security attention, thereby improving the performance of jailbreak against advanced MLLMs. Extensive experiments across different datasets and MLLMs demonstrate that the proposed MIDAS outperforms state-of-the-art jailbreak attacks for MLLMs and achieves an average attack success rate of 81.46% across 4 closed-source MLLMs. Our code is available at this [link](https://github.com/Winnie-Lian/MIDAS).

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。

↑