Think in Safety: Unveiling and Mitigating Safety Alignment Collapse in Multimodal Large Reasoning Model

TL;DR

This paper systematically evaluates the safety of 11 Multimodal Large Reasoning Models, uncovers safety degradation patterns, and proposes a safety-oriented thought process tuning approach to improve safety performance.

Contribution

It introduces a comprehensive safety evaluation framework and a novel safety-oriented tuning dataset to enhance the safety of MLRMs.

Findings

Safety degradation is prevalent across most models and benchmarks.

Long reasoning processes can improve safety performance.

Fine-tuning with the proposed dataset enhances safety on key benchmarks.

Abstract

The rapid development of Multimodal Large Reasoning Models (MLRMs) has demonstrated broad application potential, yet their safety and reliability remain critical concerns that require systematic exploration. To address this gap, we conduct a comprehensive and systematic safety evaluation of 11 MLRMs across 5 benchmarks and unveil prevalent safety degradation phenomena in most advanced models. Moreover, our analysis reveals distinct safety patterns across different benchmarks: significant safety degradation is observed across jailbreak robustness benchmarks, whereas safety-awareness benchmarks demonstrate less pronounced degradation. In particular, the long thought process in some scenarios even enhances safety performance. Therefore, it is a potential approach to address safety issues in MLRMs by leveraging the intrinsic reasoning capabilities of the model to detect unsafe intent. To operationalize this insight, we construct a multimodal tuning dataset that incorporates a safety-oriented thought process. Experimental results from fine-tuning existing MLRMs with this dataset effectively enhances the safety on both jailbreak robustness and safety-awareness benchmarks. This study provides a new perspective for developing safe MLRMs. Our dataset is available at https://github.com/xinyuelou/Think-in-Safety.