DREAM: Disentangling Risks to Enhance Safety Alignment in Multimodal Large Language Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Liu, Jianyu, Guo, Hangyu, Duan, Ranjie, Bu, Xingyuan, He, Yancheng, Li, Shilong, Huang, Hui, Liu, Jiaheng, Wang, Yucheng, Jing, Chenchen, Qu, Xingwei, Zhang, Xiao, Tan, Yingshui, Wu, Yanan, Gu, Jihao, Li, Yangguang, Zhu, Jianke
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916780175261696
author Liu, Jianyu
Guo, Hangyu
Duan, Ranjie
Bu, Xingyuan
He, Yancheng
Li, Shilong
Huang, Hui
Liu, Jiaheng
Wang, Yucheng
Jing, Chenchen
Qu, Xingwei
Zhang, Xiao
Tan, Yingshui
Wu, Yanan
Gu, Jihao
Li, Yangguang
Zhu, Jianke
author_facet Liu, Jianyu
Guo, Hangyu
Duan, Ranjie
Bu, Xingyuan
He, Yancheng
Li, Shilong
Huang, Hui
Liu, Jiaheng
Wang, Yucheng
Jing, Chenchen
Qu, Xingwei
Zhang, Xiao
Tan, Yingshui
Wu, Yanan
Gu, Jihao
Li, Yangguang
Zhu, Jianke
contents Multimodal Large Language Models (MLLMs) pose unique safety challenges due to their integration of visual and textual data, thereby introducing new dimensions of potential attacks and complex risk combinations. In this paper, we begin with a detailed analysis aimed at disentangling risks through step-by-step reasoning within multimodal inputs. We find that systematic multimodal risk disentanglement substantially enhances the risk awareness of MLLMs. Via leveraging the strong discriminative abilities of multimodal risk disentanglement, we further introduce \textbf{DREAM} (\textit{\textbf{D}isentangling \textbf{R}isks to \textbf{E}nhance Safety \textbf{A}lignment in \textbf{M}LLMs}), a novel approach that enhances safety alignment in MLLMs through supervised fine-tuning and iterative Reinforcement Learning from AI Feedback (RLAIF). Experimental results show that DREAM significantly boosts safety during both inference and training phases without compromising performance on normal tasks (namely oversafety), achieving a 16.17\% improvement in the SIUO safe\&effective score compared to GPT-4V. The data and code are available at https://github.com/Kizna1ver/DREAM.
format Preprint
id arxiv_https___arxiv_org_abs_2504_18053
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DREAM: Disentangling Risks to Enhance Safety Alignment in Multimodal Large Language Models
Liu, Jianyu
Guo, Hangyu
Duan, Ranjie
Bu, Xingyuan
He, Yancheng
Li, Shilong
Huang, Hui
Liu, Jiaheng
Wang, Yucheng
Jing, Chenchen
Qu, Xingwei
Zhang, Xiao
Tan, Yingshui
Wu, Yanan
Gu, Jihao
Li, Yangguang
Zhu, Jianke
Computation and Language
Computer Vision and Pattern Recognition
Multimodal Large Language Models (MLLMs) pose unique safety challenges due to their integration of visual and textual data, thereby introducing new dimensions of potential attacks and complex risk combinations. In this paper, we begin with a detailed analysis aimed at disentangling risks through step-by-step reasoning within multimodal inputs. We find that systematic multimodal risk disentanglement substantially enhances the risk awareness of MLLMs. Via leveraging the strong discriminative abilities of multimodal risk disentanglement, we further introduce \textbf{DREAM} (\textit{\textbf{D}isentangling \textbf{R}isks to \textbf{E}nhance Safety \textbf{A}lignment in \textbf{M}LLMs}), a novel approach that enhances safety alignment in MLLMs through supervised fine-tuning and iterative Reinforcement Learning from AI Feedback (RLAIF). Experimental results show that DREAM significantly boosts safety during both inference and training phases without compromising performance on normal tasks (namely oversafety), achieving a 16.17\% improvement in the SIUO safe\&effective score compared to GPT-4V. The data and code are available at https://github.com/Kizna1ver/DREAM.
title DREAM: Disentangling Risks to Enhance Safety Alignment in Multimodal Large Language Models
topic Computation and Language
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.18053