When One Modality Rules Them All: Backdoor Modality Collapse in Multimodal Diffusion Models

Multimodal Diffusion Language Models (MDLMs) have recently emerged as a competitive alternative to their autoregressive counterparts. Yet their vulnerability to backdoor attacks remains largely unexplored. In this work, we show that well-established data-poisoning pipelines can successfully implant backdoors into MDLMs, enabling attackers to manipulate model behavior via specific triggers while maintaining normal performance on clean inputs. However, defense strategies effective to these models are yet to emerge. To bridge this gap, we introduce a backdoor defense framework for MDLMs named DiSP (Diffusion Self-Purification). DiSP is driven by a key observation: selectively masking certain vision tokens at inference time can neutralize a backdoored model's trigger-induced behaviors and restore normal functionality. Building on this, we purify the poisoned dataset using the compromised model itself, then fine-tune the model on the purified data to recover it to a clean one. Given such a specific design, DiSP can remove backdoors without requiring any auxiliary models or clean reference data. Extensive experiments demonstrate that our approach effectively mitigates backdoor effects, reducing the attack success rate (ASR) from over 90% to typically under 5%, while maintaining model performance on benign tasks.

Backdoor attacks undermine the reliability and trustworthiness of machine learning systems by injecting hidden behaviors that can be maliciously activated at inference time. While such threats have been extensively studied in unimodal settings, their impact on multimodal foundation models, particularly vision-language models (VLMs), remains largely underexplored. In this work, we introduce BackdoorVLM, the first comprehensive benchmark for evaluating backdoor attacks and defenses on VLMs across a broad range of settings. It adopts a unified perspective that injects and analyzes backdoors across core vision-language tasks, including image captioning and visual question answering. BackdoorVLM organizes multimodal backdoor threats into 5 representative categories: targeted refusal, malicious injection, jailbreak, concept substitution, and perceptual hijack. Each category captures a distinct pathway through which an adversary can manipulate a model's behavior. We evaluate these threats using 12 representative attack methods spanning text, image, and bimodal triggers, tested on 2 open-source VLMs and 3 multimodal datasets along with 5 representative defense methods. Our analysis reveals that VLMs exhibit strong sensitivity to textual instructions, and in bimodal backdoors the text trigger typically overwhelms the image trigger when forming the backdoor mapping. Notably, backdoors involving the textual modality remain highly potent, with poisoning rates as low as 1% yielding over 90% success across most tasks. Moreover, existing defenses lack generalization across trigger types and backdoor scenarios, leading to unreliable protection against diverse triggers across modalities and forms. These findings highlight significant, underexplored vulnerabilities in current VLMs. We hope that BackdoorVLM can serve as a useful benchmark for analyzing and mitigating multimodal backdoor threats.

Multimodal Large Language Models (MLLMs) are increasingly deployed in safety-critical domains, yet their susceptibility to backdoor attacks remains a formidable challenge. While existing defenses have evolved from visual-centric patch cues to unsupervised internal attention statistics, they are fundamentally constrained by a reliance on static snapshots and attack-specific heuristics. Given the intricate interplay between dual modalities, the vast design space for multimodal triggers allows diverse threats, such as textual injections, sub-perceptual noises, and coordinated cross-modal triggers, to easily circumvent such rigid priors, particularly when attack types are unknown in real-world applications. In this paper, we rethink backdoor detection by shifting the diagnostic paradigm from static observations to dynamic output evolution. We observe that a successful attack inevitably hijacks the model's decoding trajectory, skewing internal modal contributions during generation while maintaining an abnormal output invariance despite input perturbations. To capture this behavior, we propose EVO-Detect, a unified detection framework that monitors the dynamic evolution of internal signals during early decoding steps through two complementary channels: Attention Flow and Logits Flow. The attention branch measures how image-text attention allocation evolves during early generation, whereas the logits branch quantifies the sensitivity of the output distribution to modal perturbations through interventional ablations. We then convert these trajectories into compact sample-level metrics and normalize them against negative-control signatures so that natural perturbations are less likely to be mistaken for backdoor evidence. Extensive evaluations show that our approach substantially outperforms state-of-the-art baselines such as BYE and TCAP across diverse attack families.