SafeMol: Dual-Modality Safety Alignment for Molecular Multimodal Models

2026-09-27

AI Summary

This study addresses the safety vulnerabilities and insufficient cross-modal robustness of molecular multimodal models in hazardous molecule scenarios. To this end, it introduces SafeMolBench, a pioneering safety alignment benchmark, and proposes a parameter-efficient multimodal alignment framework. Methodologically, the framework achieves distribution-level cross-modal representation alignment via Maximum Mean Discrepancy, synergistically enhancing model safety through explicit harmful intent modeling and joint text-graph optimization. Experimental results demonstrate that the proposed approach significantly reduces attack success rates while maintaining low over-refusal rates and effectively preserving utility across downstream molecular tasks, thereby achieving an optimal balance among safety, reliability, and practicality.

Abstract

Molecular multimodal models support diverse understanding and generation tasks but may introduce safety vulnerabilities when handling hazardous molecules. In this work, We reveal substantial jailbreak vulnerabilities under both text-only and graph-conditioned settings. Our analysis further shows that safety robustness must hold across input modalities while balancing safety, over-refusal, and utility. To address these challenges, we construct SafeMolBench, a molecular multimodal safety-alignment benchmark with 3702 samples covering 618 unique hazardous molecules and safe molecular tasks, organized into hazardous-harmful, hazardous-allowed, and utility-replay subsets to support unified training and evaluation of safety, over-refusal, and utility. Based on SafeMolBench, we propose SafeMol, a parameter-efficient safety alignment framework that jointly optimizes lightweight modules across text-only and graph-conditioned inputs, uses MMD for distribution-level representation alignment to reduce modality-induced discrepancies, and explicitly models molecular hazardousness and harmful operational intent. Experiments on SafeMolBench show that SafeMol reduces attack success by several tens of percentage points while largely maintaining low over-refusal and preserving molecular-task utility.

Problem

• molecular multimodal models
• safety alignment
• jailbreak vulnerabilities
• over-refusal
• hazardous molecules

Innovation

• Safety Alignment
• Molecular Multimodal Models
• Parameter-Efficient Fine-Tuning
• Maximum Mean Discrepancy
• Jailbreak Vulnerability