LLaVAShield：为视觉语言模型中的多模态多轮对话提供安全保障
LLaVAShield: Safeguarding Multimodal Multi-Turn Dialogues in Vision-Language Models
为视觉语言模型提供多模态多轮对话的安全审核器与 MMDS 基准，可直接用于评测和加固 VLM 在交互式多轮场景下的输入与回复安全性，并提供 MMRT 自动化红队生成框架来构造对抗性多模态对话。
原文依据 we first construct the Multimodal Multi-turn Dialogue Safety (MMDS) dataset, comprising 4,484 annotated dialogues
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论随着视觉语言模型（VLM）进入交互式、多轮使用场景，多模态多轮对话的安全隐患日益突出，其特点是恶意意图具有隐蔽性、上下文风险会逐步累积，以及跨模态联合风险。这些特点限制了为单轮或单模态场景设计的内容审核方法的有效性。为应对这些局限，我们首先构建了多模态多轮对话安全（MMDS）数据集，包含 4,484 条带标注对话，以及一套涵盖 8 个一级维度和 60 个二级维度的完整风险分类体系。在 MMDS 的构建过程中，我们提出了多模态多轮红队测试（MMRT），这是一个用于生成不安全多模态多轮对话的自动化框架。我们进一步提出 LLaVAShield，它在多模态多轮对话中按指定的策略维度审核用户输入与助手回复的安全性。大量实验表明，LLaVAShield 显著优于最优的 VLM 和现有内容审核工具，同时展现出较强的泛化能力和灵活的策略适配能力。此外，我们分析了主流 VLM 对有害输入的脆弱性，并评估了关键组件的贡献，从而推进了对多模态多轮对话中安全机制的理解。
英文原文摘要 1343 字符
As Vision-Language Models (VLMs) move into interactive, multi-turn use, safety concerns intensify for multimodal multi-turn dialogue, which is characterized by concealment of malicious intent, contextual risk accumulation, and cross-modal joint risk. These characteristics limit the effectiveness of content moderation approaches designed for single-turn or single-modality settings. To address these limitations, we first construct the Multimodal Multi-turn Dialogue Safety (MMDS) dataset, comprising 4,484 annotated dialogues and a comprehensive risk taxonomy with 8 primary and 60 subdimensions. As part of MMDS construction, we introduce Multimodal Multi-turn Red Teaming (MMRT), an automated framework for generating unsafe multimodal multi-turn dialogues. We further propose LLaVAShield, which audits the safety of both user inputs and assistant responses under specified policy dimensions in multimodal multi-turn dialogues. Extensive experiments show that LLaVAShield significantly outperforms state-of-the-art VLMs and existing content moderation tools while demonstrating strong generalization and flexible policy adaptation. Additionally, we analyze vulnerabilities of mainstream VLMs to harmful inputs and evaluate the contribution of key components, advancing understanding of safety mechanisms in multimodal multi-turn dialogues.