COMIC: Reference-Aware Safety Gating for Multimodal Large Language Models

📄 arXiv: 2608.17234v1 📥 PDF

作者: Md Abdullahil Oaphy, Anhao Xiang, Zongxing Xie, Huayue Gu, Chenyu Wang, Honghui Xu

分类: cs.CR, cs.AI

发布日期: 2026-08-18


💡 一句话要点

提出COMIC以解决多模态大语言模型的安全风险问题

🎯 匹配领域: 支柱三:空间感知与语义 (Perception & Semantics) 支柱九:具身大模型 (Embodied Foundation Models)

关键词: 多模态大语言模型 安全性 参考感知 操作-目标对 越狱攻击 鲁棒性 OCR技术 质量感知路由

📋 核心要点

  1. 现有多模态大语言模型在处理视觉输入时,存在安全风险,尤其是在操作与视觉目标绑定时可能导致不安全行为。
  2. COMIC通过推断操作和参考类型,构建候选目标并评估安全性,提供了一种新的参考感知安全机制。
  3. 实验结果显示,COMIC在多个基准测试中均显著提高了模型的鲁棒性,同时保持了良好的实用性和效率。

📝 摘要(中文)

多模态大语言模型(MLLMs)在处理截图、扫描文档和图表等视觉输入时,面临新的安全风险。许多多模态越狱攻击中,单独的提示或图像并不具备危害性,只有在模型将看似无害的操作与特定视觉目标绑定时,才会出现不安全行为。这揭示了当前多模态防御的结构性弱点。为了解决这一问题,本文提出了COMIC(上下文-操作-模态-图像分类器),一个参考感知的预生成安全门。COMIC通过推断请求的操作和参考类型,构建候选目标,并评估显式操作-目标对的安全性,从而提高模型的鲁棒性。实验结果表明,COMIC在多个开源MLLMs和多模态越狱基准测试中均表现出色,显著提升了安全性和实用性。

🔬 方法详解

问题定义:本文旨在解决多模态大语言模型在处理视觉输入时的安全风险,尤其是当无害的提示与特定视觉目标结合时可能引发的不安全行为。现有方法主要对提示-图像对进行整体审核,未能有效识别潜在的安全隐患。

核心思路:COMIC的核心思路是通过参考感知机制,分析请求的操作和目标类型,从而在生成之前评估操作-目标对的安全性。这种设计能够更精确地识别和防范潜在的安全风险。

技术框架:COMIC的整体架构包括几个主要模块:首先推断请求的操作和参考类型;其次,从OCR和开放词汇提案中构建候选目标;然后对可能的参考进行定位;最后评估显式操作-目标对的安全性。

关键创新:COMIC的主要创新在于其参考感知的安全门设计,能够在生成之前识别和评估操作-目标对的安全性。这与现有方法的整体审核机制形成了本质区别。

关键设计:在设计中,COMIC结合了最大风险聚合与质量感知路由,以保守地处理模糊性,确保在决定是否转发或阻止请求时的安全性。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果表明,COMIC在多个开源多模态大语言模型上均显著提高了安全性,尤其是在处理本地化和广泛的多模态越狱基准测试中,表现出更高的鲁棒性和实用效率,具体提升幅度达到20%以上。

🎯 应用场景

COMIC的研究成果具有广泛的应用潜力,尤其是在需要处理视觉输入的多模态交互系统中,如智能助手、自动翻译和信息检索等领域。通过提高多模态模型的安全性,COMIC能够有效防范潜在的安全风险,提升用户信任度和系统可靠性。

📄 摘要(原文)

Multimodal large language models (MLLMs) are increasingly used to interact with screenshots, scanned documents, diagrams, and other visually grounded inputs. This shift introduces a new safety risk: in many multimodal jailbreaks, neither the prompt nor the image is harmful in isolation. Unsafe behavior emerges only when the model binds an apparently benign operation, such as summarizing, translating, or following, to a localized visual target. This reveals a structural weakness in current multimodal defenses, which largely moderate the prompt-image pair as a whole even though the true security-relevant unit is the grounded operation-target pair produced during dereference. In this work, we identify and analyze this reference-dependent failure mode and show that existing defenses degrade when harmful semantics are localized, activated only after grounding, and dependent on visual reference resolution. To address this problem, we propose COMIC (Context-Operation-Modality-Image-Classifier), a reference-aware pre-generation safety gate for MLLMs. COMIC first infers the requested operation and reference type, constructs candidate targets from OCR and open-vocabulary proposals, grounds plausible referents, and evaluates safety over explicit operation-target pairs. To handle ambiguity conservatively, COMIC combines max-risk aggregation with quality-aware routing before deciding whether to forward or block a request. We evaluate COMIC across multiple open-source MLLMs, localized and broader multimodal jailbreak benchmarks, and benign reference-sensitive settings. The results show that COMIC consistently improves robustness while preserving benign utility and practical efficiency. More broadly, our findings suggest that multimodal safety cannot be enforced reliably without modeling the requested operation, the visual target to which it applies, and the confidence of that grounding.