Securing RAG Pipelines with Fine Grained Authorization
summary
Shows a concrete ReBAC implementation using SpiceDB with Pinecone embeddings, OpenAI for the LLM and LangChain to orchestrate; demonstrates a pre-filter flow where a fake document (“Bill Gates won the 2025 Oscar for best football movie”) is upserted, Tim is granted viewer relation, the LLM returns that fact, the relation is deleted, and subsequent queries return nothing. Claims fine‑grained authorization, audit logging, and improved query latency; cites Zanzibar’s 10M QPS scale and notes pre vs post filtering tradeoffs depending on vector hit rate.
AI-written summary · about
chapters (8)
- 0:00Securing RAG pipelines
- 1:57Authorization for RAG
- 5:47History of authorization models
- 8:58Relationship-based access control
- 12:55Fine-grained authorization for RAG
- 15:26Pre-filter vs post-filter
- 18:20Live demo: authorization in RAG
- 27:17Q&A: SpiceDB and scalability
related talks
40mSecuring AI RAG Pipelines with Fine Grained Authorization
44mSecuring RAG Pipelines with Fine Grained Authorization
39mSecuring RAG Pipelines with Fine Grained Authorization
33mHow to Prevent AI Agents from Accessing Unauthorized Data
30mFine-Grained Authorization: The Missing Piece in Agentic AI Security
57mFine-Grained Authorization: The Missing Piece in Agentic AI Security
55mYour AI Agents Are Overprivileged: The Case for Fine-Grained Authorization
47mFine Grained Authorisation with Relationship Based Access Control