RAG Databases
A RAG database is like a digital library connected to an AI. Because the AI can look up information in this library to answer questions, a malicious user can trick the AI into reading and revealing almost any private document stored inside it.
A RAG database acts as an external knowledge base for an AI model. Security risks arise when access controls are insufficient, allowing an adversary to craft prompts that force the AI to retrieve and output sensitive, unauthorized internal documents from the database.
A Retrieval-Augmented Generation (RAG) database serves as a vector-indexed retrieval system for LLM context injection. Adversarial prompt engineering exploits the lack of granular authorization at the retrieval layer, enabling unauthorized data exfiltration of the organization's indexed corpus through the model's generation interface.