RAG Security and Privacy: Formalizing the Threat Model and Attack Surface
A new paper introduces the first formal threat model for Retrieval-Augmented Generation (RAG) systems, addressing critical privacy and security gaps. The research defines a taxonomy of adversaries and formalizes attack vectors like document-level membership inference and data poisoning. This work aims to provide a more rigorous understanding of security and privacy risks inherent in RAG deployments. AI
IMPACT Establishes a foundational framework for understanding and mitigating security and privacy risks in RAG systems.