Beyond Text: Unveiling Privacy Vulnerabilities in Multi-modal Retrieval-Augmented Generation
This addresses privacy risks for users of multimodal AI systems, which is an incremental but important extension of prior text-based RAG studies.
The paper tackles privacy vulnerabilities in multimodal retrieval-augmented generation (MRAG) systems, demonstrating through a novel compositional structured prompt attack that attackers can extract private information from vision-language and speech-language data, with experiments showing LMMs can directly generate retrieved content or indirectly expose sensitive details.
Multimodal Retrieval-Augmented Generation (MRAG) systems enhance LMMs by integrating external multimodal databases, but introduce unexplored privacy vulnerabilities. While text-based RAG privacy risks have been studied, multimodal data presents unique challenges. We provide the first systematic analysis of MRAG privacy vulnerabilities across vision-language and speech-language modalities. Using a novel compositional structured prompt attack in a black-box setting, we demonstrate how attackers can extract private information by manipulating queries. Our experiments reveal that LMMs can both directly generate outputs resembling retrieved content and produce descriptions that indirectly expose sensitive information, highlighting the urgent need for robust privacy-preserving MRAG techniques.