Definition
In RAG and AI Agent architectures, Access Control refers to the enforcement of document-level security (DLS) or row-level permissions within a vector database to ensure retrieved context fragments align with a user's specific authorization. This prevents 'data leakage' where an LLM generates answers based on sensitive information the querying user is not permitted to see.
Focuses on granular document-level authorization within the data retrieval layer rather than simple user login (Authentication).
"A keycard system for a library where the reader can see the building, but only specific bookshelves unlock based on their security clearance level."
Conceptual Overview
In RAG and AI Agent architectures, Access Control refers to the enforcement of document-level security (DLS) or row-level permissions within a vector database to ensure retrieved context fragments align with a user's specific authorization. This prevents 'data leakage' where an LLM generates answers based on sensitive information the querying user is not permitted to see.
Disambiguation
Focuses on granular document-level authorization within the data retrieval layer rather than simple user login (Authentication).
Visual Analog
A keycard system for a library where the reader can see the building, but only specific bookshelves unlock based on their security clearance level.