Medusa: Cross-Modal Transferable Adversarial Attacks on Multimodal Medical Retrieval-Augmented Generation
Open Access
- 20 April 2026
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 1275-1286
Abstract
With the rapid advancement of retrieval-augmented vision-language models, multimodal medical retrieval-augmented generation (MMed-RAG) systems are increasingly adopted in clinical decision support. These systems enhance medical applications by performing cross-modal retrieval to integrate relevant visual and textual evidence for tasks, e.g., report generation and disease diagnosis. However, their complex architecture also introduces underexplored adversarial vulnerabilities, particularly via visual input perturbations. In this paper, we propose Medusa, a novel framework for crafting cross-modal transferable adversarial attacks on MMed-RAG systems under a black-box setting. Specifically, Medusa formulates the attack as a perturbation optimization problem, leveraging a multi-positive InfoNCE loss (MPIL) to align adversarial visual embeddings with medically plausible but malicious textual targets, thereby hijacking the retrieval process. To enhance transferability, we adopt a surrogate model ensemble and design a dual-loop optimization strategy augmented with invariant risk minimization (IRM). Extensive experiments on two real-world medical tasks, including medical report generation and disease diagnosis, demonstrate that Medusa achieves over 90% average attack success rate across various generation models and retrievers under appropriate parameter configuration, while remaining robust against four mainstream defenses, outperforming state-of-the-art baselines. Our results reveal critical vulnerabilities in the MMed-RAG systems and highlight the necessity of robustness benchmarking in safety-critical medical applications. The code and data are available at https://github.com/yiliucs/MMed-RAG-Attack.Keywords
All Related Versions
Funding Information
- Zhejiang Leading Innovative and Entrepreneur Team Introduction Program (2024R01007)
This publication has 29 references indexed in Scilit:
- ChatCAD: An MLLM-Guided Framework for Zero-shot CAD Drawing Restoration2025
- Medical large language models are susceptible to targeted misinformation attacksnpj Digital Medicine, 2024
- MediCLIP: Adapting CLIP for Few-Shot Medical Image Anomaly Detection2024
- Transferable Multimodal Attack on Vision-Language Pre-training Models2024
- Transferable Decoding with Visual Entities for Zero-Shot Image Captioning2023
- PMC-CLIP: Contrastive Language-Image Pre-training Using Biomedical Documents2023
- Joint Learning of Localized Representations from Medical Images and Reports2022
- Stochastic Variance Reduced Ensemble Adversarial Attack for Boosting the Adversarial Transferability2022
- MedRDF: A Robust and Retrain-Less Diagnostic Framework for Medical Pretrained Models Against Adversarial AttackIEEE Transactions on Medical Imaging, 2022
- ComDefend: An Efficient Image Compression Model to Defend Adversarial Examples2019