When One Modality Rules Them All: Backdoor Modality Collapse in Multimodal Diffusion Models
2026-03-06 • Machine Learning
Machine Learning
AI summaryⓘ
The authors studied how backdoor attacks work in models that process multiple types of information at once, like text and images. They found that instead of all parts working together, usually one type of information ends up controlling the attack, making the other parts unnecessary. To understand this better, the authors made two new ways to measure how much each type of information contributes to the attack. Their experiments showed that the expected teamwork between different types doesn’t happen, which could make it easier to spot and defend against these attacks in the future.
diffusion modelsbackdoor attacksmultimodal modelstrigger modality attributioncross-trigger interactionconditional diffusionmodel vulnerabilitymechanistic analysis
Authors
Qitong Wang, Haoran Dai, Haotian Zhang, Christopher Rasmussen, Binghui Wang
Abstract
While diffusion models have revolutionized visual content generation, their rapid adoption has underscored the critical need to investigate vulnerabilities, e.g., to backdoor attacks. In multimodal diffusion models, it is natural to expect that attacking multiple modalities simultaneously (e.g., text and image) would yield complementary effects and strengthen the overall backdoor. In this paper, we challenge this assumption by investigating the phenomenon of Backdoor Modality Collapse, a scenario where the backdoor mechanism degenerates to rely predominantly on a subset of modalities, rendering others redundant. To rigorously quantify this behavior, we introduce two novel metrics: Trigger Modality Attribution (TMA) and Cross-Trigger Interaction (CTI). Through extensive experiments across diverse training configurations in multimodal conditional diffusion, we consistently observe a ``winner-takes-all'' dynamic in backdoor behavior. Our results reveal that (1) attacks often collapse into subset-modality dominance, and (2) cross-modal interaction is negligible or even negative, contradicting the intuition of synergistic vulnerability. These findings highlight a critical blind spot in current assessments, suggesting that high attack success rates often mask a fundamental reliance on a subset of modalities. This establishes a principled foundation for mechanistic analysis and future defense development.