Problem
Research questions and friction points this paper is trying to address.
Unified Multimodal Models
safety
generation-understanding coupling
cross-functionality reciprocity
multimodal security
Unsafe by Reciprocity: How Generation-Understanding Coupling Undermines Safety in Unified Multimodal Models
🤖 AI Summary
This work addresses the underexplored structural security risks arising from the tight coupling of generation and understanding capabilities in unified multimodal models. Treating their reciprocal integration as a potential vulnerability, we propose the RICE attack paradigm, which exploits bidirectional cross-functional interactions—both generation-to-understanding and understanding-to-generation—to uncover propagation pathways of unsafe cross-modal signals. We introduce an intermediate signal analysis method grounded in shared architectures and develop a multimodal safety evaluation framework. Empirical evaluations across multiple state-of-the-art unified models demonstrate high attack success rates, providing the first evidence that both interaction directions harbor significant security flaws. Our findings offer a novel perspective and a foundational benchmark for assessing safety in multimodal systems.
📝 Abstract
Recent advances in Large Language Models (LLMs) and Text-to-Image (T2I) models have led to the emergence of Unified Multimodal Models (UMMs), where multimodal understanding and image generation are tightly integrated within a shared architecture. Prior studies suggest that such reciprocity enhances cross-functionality performance through shared representations and joint optimization. However, the safety implications of this tight coupling remain largely unexplored, as existing safety research predominantly analyzes understanding and generation functionalities in isolation. In this work, we investigate whether cross-functionality reciprocity itself constitutes a structural source of vulnerability in UMMs. We propose RICE: Reciprocal Interaction-based Cross-functionality Exploitation, a novel attack paradigm that explicitly exploits bidirectional interactions between understanding and generation. Using this framework, we systematically evaluate Generation-to-Understanding (G-U) and Understanding-to-Generation (U-G) attack pathways, demonstrating that unsafe intermediate signals can propagate across modalities and amplify safety risks. Extensive experiments show high Attack Success Rates (ASR) in both directions, revealing previously overlooked safety weaknesses inherent to UMMs.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Unified Multimodal Models
Cross-functionality Reciprocity
RICE Attack
Safety Vulnerability
Bidirectional Interaction
🔎 Similar Papers
No similar papers found.
