连贯性陷阱：当多模态大语言模型构建的叙事利用被操纵的视觉上下文
The Coherence Trap: When MLLM-Crafted Narratives Exploit Manipulated Visual Contexts
提出利用MLLM生成语义连贯的虚假图文对来构建对抗数据集，并设计伪影感知编码与操作导向推理框架，为检测MLLM驱动的多模态虚假信息提供了可复用的数据生成与检测思路。
原文依据 we propose a new adversarial pipeline that leverages MLLMs to generate high-risk disinformation
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媒体操纵的检测与定位已成为对抗AI生成虚假信息的关键挑战。尽管现有方法近年来取得了进展，但我们发现当前方法存在两个根本性局限：（1）低估了多模态大语言模型驱动的欺骗风险：主流技术主要处理基于规则的文本操纵，却未能考虑由多模态大语言模型合成的复杂虚假信息，这些模型能够根据被操纵的图像动态生成语义连贯、上下文合理但具有欺骗性的叙事；（2）不现实的错位伪影：当前关注的场景依赖于缺乏语义连贯性的人工错位内容，使其易于检测。为了全面解决这些差距，我们提出了一种新的对抗流程，利用多模态大语言模型生成高风险的虚假信息。我们的方法首先构建多模态大语言模型驱动的合成多模态数据集，其中图像首先使用最优的编辑技术进行修改，然后与多模态大语言模型生成的欺骗性文本配对，这些文本与视觉操纵保持语义一致性。在此基础上，我们提出了通过多模态大语言模型进行伪影感知操纵诊断的框架，该框架具有两项关键创新：伪影预感知编码策略和面向操纵的推理，以驯服多模态大语言模型解决该数据集问题。全面的实验验证了我们的框架作为检测多模态大语言模型驱动的多模态欺骗的统一架构具有优越的泛化能力。在该数据集上的跨域测试中，该框架取得了最佳的平均性能，准确率、平均精度均值和平均交并比分别为88.18、60.25和61.02。
英文原文摘要 1789 字符
The detection and grounding of multimedia manipulation has emerged as a critical challenge in combating AI-generated disinformation. While existing methods have made progress in recent years, we identify two fundamental limitations in current approaches: (1) Underestimation of MLLM-driven deception risk: prevailing techniques primarily address rule-based text manipulations, yet fail to account for sophisticated misinformation synthesized by multimodal large language models (MLLMs) that can dynamically generate semantically coherent, contextually plausible yet deceptive narratives conditioned on manipulated images; (2) Unrealistic misalignment artifacts: currently focused scenarios rely on artificially misaligned content that lacks semantic coherence, rendering them easily detectable. To address these gaps holistically, we propose a new adversarial pipeline that leverages MLLMs to generate high-risk disinformation. Our approach begins with constructing the MLLM-Driven Synthetic Multimodal (MDSM) dataset, where images are first altered using state-of-the-art editing techniques and then paired with MLLM-generated deceptive texts that maintain semantic consistency with the visual manipulations. Building upon this foundation, we present the Artifact-aware Manipulation Diagnosis via MLLM (AMD) framework featuring two key innovations: Artifact Pre-perception Encoding strategy and Manipulation-Oriented Reasoning, to tame MLLMs for the MDSM problem. Comprehensive experiments validate our framework's superior generalization capabilities as a unified architecture for detecting MLLM-powered multimodal deceptions. In cross-domain testing on the MDSM dataset, AMD achieves the best average performance, with ACC, mAP, and mIoU scores of 88.18, 60.25, and 61.02, respectively.