MSR-Align: Policy-Grounded Multimodal Alignment for Safety-Aware Reasoning in Vision-Language Models
This addresses safety alignment for reasoning-capable VLMs, which is a domain-specific problem, but the approach is incremental as it builds on existing safety datasets and fine-tuning methods.
The paper tackles the problem of safety risks in vision-language models (VLMs) from harmful multimodal prompts by introducing MSR-Align, a high-quality multimodal safety reasoning dataset, and shows that fine-tuning VLMs on it substantially improves robustness against jailbreak attacks while preserving general reasoning performance.
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.