Securing AI RAG Pipelines with Fine Grained Authorization

Learn how to implement fine-grained authorization in AI Retrieval-Augmented Generation (RAG) pipelines using relationship-based access control (ReBAC) in this 40-minute conference talk from NDC Oslo 2025. Discover why traditional authorization models fall short for enterprise AI applications and explore how ReBAC provides more precise control by making decisions based on relationships between objects. Understand the critical importance of authorization in RAG pipelines and examine how Google Zanzibar achieves secure data access through ReBAC principles. Explore techniques for pre-filtering vector database queries with authorized object IDs to enhance both efficiency and security. Follow along with a practical demonstration that implements fine-grained authorization for RAG systems using Pinecone, Langchain, OpenAI, and SpiceDB, providing you with concrete tools and strategies to ensure users can only access data they're authorized to use when augmenting AI prompts.