MSR-Align：面向视觉-语言模型安全感知推理的基于策略的多模态对齐
MSR-Align: Policy-Grounded Multimodal Alignment for Safety-Aware Reasoning in Vision-Language Models
AI总结：
针对现有VLM安全对齐方法及数据集的不足，本研究提出高质量多模态安全推理数据集MSR-Align，实验证明其微调可提升VLM抗越狱鲁棒性且保留推理性能。
摘要：
视觉-语言模型（VLM）通过增强思维链能力，已在多模态推理任务中取得显著进展。然而，这一进展也带来了新的安全风险，因为这类模型越来越容易受到有害多模态提示的攻击，进而引发不道德或不安全的行为。现有的安全对齐方法主要是为单模态语言模型设计的，不足以应对多模态输入带来的复杂且隐蔽的威胁。此外，当前的安全数据集缺乏细粒度的、基于策略的推理能力，而这正是对具备推理能力的VLM进行鲁棒对齐所必需的。在本研究中，我们提出了MSR-Align，这是一个专为弥补这一缺口而构建的高质量多模态安全推理数据集。MSR-Align支持跨视觉和文本模态、基于标准化安全策略的细粒度审慎推理。我们的数据生成流程强调多模态多样性、基于策略的推理，并采用强大的多模态评判器进行严格的质量过滤。大量实验表明，在MSR-Align上微调VLM，可显著提升模型针对文本和视觉-语言越狱攻击的鲁棒性，同时保持甚至提升通用推理性能。MSR-Align为推进具备推理能力的VLM的安全对齐提供了可扩展且有效的基础。我们的数据集已公开发布于https://huggingface.co/datasets/Leigest/MSR-Align。
英文摘要：
Vision-Language Models (VLMs) have achieved remarkable progress in multimodal reasoning tasks through enhanced chain-of-thought capabilities. However, this advancement also introduces novel safety risks, as these models become increasingly vulnerable to harmful multimodal prompts that can trigger unethical or unsafe behaviors. Existing safety alignment approaches, primarily designed for unimodal language models, fall short in addressing the complex and nuanced threats posed by multimodal inputs. Moreover, current safety datasets lack the fine-grained, policy-grounded reasoning required to robustly align reasoning-capable VLMs. In this work, we introduce {MSR-Align}, a high-quality Multimodal Safety Reasoning dataset tailored to bridge this gap. MSR-Align supports fine-grained, deliberative reasoning over standardized safety policies across both vision and text modalities. Our data generation pipeline emphasizes multimodal diversity, policy-grounded reasoning, and rigorous quality filtering using strong multimodal judges. Extensive experiments demonstrate that fine-tuning VLMs on MSR-Align substantially improves robustness against both textual and vision-language jailbreak attacks, while preserving or enhancing general reasoning performance. MSR-Align provides a scalable and effective foundation for advancing the safety alignment of reasoning-capable VLMs. Our dataset is made publicly available at https://huggingface.co/datasets/Leigest/MSR-Align.
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。
