RapGuard: Safeguarding Multimodal Large Language Models via Rationale-aware Defensive Prompting

2024/12/25 by Yilei Jiang, Jiang, Yilei, Y. H. Tan +3 · 4 citations
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2412.18826

openalex publication_date 2024/12/25 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

While Multimodal Large Language Models (MLLMs) have made remarkable progress in vision-language reasoning, they are also more susceptible to producing harmful content compared to models that focus solely on text. Existing defensive prompting techniques rely on a static, unified safety guideline that fails to account for the specific risks inherent in different multimodal contexts. To address these limitations, we propose RapGuard, a novel framework that uses multimodal chain-of-thought reasoning to dynamically generate scenario-specific safety prompts. RapGuard enhances safety by adapting its prompts to the unique risks of each input, effectively mitigating harmful outputs while maintaining high performance on benign tasks. Our experimental results across multiple MLLM benchmarks demonstrate that RapGuard achieves state-of-the-art safety performance, significantly reducing harmful content without degrading the quality of responses.

Cited by

• QuadSentinel: Sequent Safety for Machine-Checkable Control in Multi-agent Systems
• SaFeR-VLM: Toward Safety-aware Fine-grained Reasoning in Multimodal Models
• LLaVAShield: Safeguarding Multimodal Multi-Turn Dialogues in Vision-Language Models
• A Survey on Training-free Alignment of Large Language Models

Related

• VLMGuard-R1: Proactive Safety Alignment for VLMs via Reasoning-Driven Prompt Optimization
• Automating Steering for Safe Multimodal Large Language Models
• MSR-Align: Policy-Grounded Multimodal Alignment for Safety-Aware Reasoning in Vision-Language Models
• Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs
• GuardTrace-VL: Detecting Unsafe Multimodel Reasoning via Iterative Safety Supervision