Walking the Embedding Space: Datastore Extraction from Multimodal RAG

2026-10-01

AI Summary

This study addresses the privacy leakage risks in multimodal Retrieval-Augmented Generation (RAG) systems arising from data extraction attacks. We propose Immrag, a black-box attack framework that innovatively embeds malicious instructions into input images rather than text prompts, enabling non-interactive, automated data extraction. The method integrates a CLIP retriever, relevance-weighted resampling, and adversarial image processing to efficiently extract visual data through adaptive traversal of the embedding space. Experimental results demonstrate that in sensitive domains such as healthcare, a single execution can reconstruct hundreds of images, achieving an extraction efficiency 5.6 times higher than baseline methods. These findings reveal significant security vulnerabilities inherent in multimodal RAG architectures.

Abstract

Multimodal Retrieval-Augmented Generation (MRAG) has emerged as a reliable and cost-effective technique of grounding the generative capabilities of Multimodal Large Language Models (MLLMs) into relevant, up-to-date, external knowledge. Despite presenting several benefits, such as reducing hallucinatory behavior, they also introduce new attack surfaces, including leakage of private information and vulnerabilities against data extraction attacks. In this paper, we introduce $\immrag$, an adaptive and automatic data extraction attack procedure operating in a black box setting against \emph{image-returning} MRAG, a configuration in which the retrieved visual artifact is itself the response. Each query blends an attacker-held shadow image with an image already recovered from the system, and relevance-weighted resampling steers subsequent queries towards regions of the embedding space that still yield novel retrievals. Unlike current extraction attacks that aim to persuade the model towards data leakage by placing a malicious query as a textual prompt, $\immrag$ embeds the malicious instructions inside a user-given input image. We evaluate $\immrag$ on three plausible and distinct real-world scenarios: medical assistant, document-focused helper and general purpose tool. The experiments involve the study of the effectiveness of the attack on multiple CLIP-family retrievers, as well as the impact of various generators. A single 2500-query run reconstructs up to 611 distinct radiology images, 566 document scans and 416 general-purpose images under local-feature correspondence, and reaches up to $5.6\times$ as many distinct datastore items as a non-adaptive baseline. Our results show the urgent need for safeguards specifically designed for multimodal data.

Problem

• Multimodal RAG
• Datastore Extraction
• Black-box Attack
• Data Leakage
• Security Vulnerability

Innovation

• Multimodal RAG
• Data Extraction Attack
• Black-box Attack
• Embedding Space Exploration
• Adversarial Image