Securing RAG Pipelines with Fine Grained Authorization
summary
Claims that Relationship‑based Access Control (Zanzibar/ReBAC) prevents RAG leaks by pre-filtering vector DB queries with authorized object IDs and by post‑filtering results are demonstrated end to end. The demo wires SpiceDB + Pinecone + Langchain + OpenAI, shows deleting a relation immediately removes sensitive content from responses, and cites Google Zanzibar’s 10M qps claim plus a real deployment handling 37 billion documents across 5 million users. Implementation details, latency tradeoffs, and when to use pre‑ vs post‑filtering are shown and validated.
AI-written summary · about
chapters (8)
- 0:00Authorization challenges
- 4:45Google Zanzibar and relationship-based access
- 12:56RAG pipelines and AI security
- 17:13Open-source Zanzibar implementations
- 21:08Pre-filter vs post-filter authorization
- 24:01Fine-grained authorization demo
- 32:29Zanzibar at OpenAI production
- 35:20Q&A
related talks
44mSecuring RAG Pipelines with Fine Grained Authorization
33mSecuring RAG Pipelines with Fine Grained Authorization
40mSecuring AI RAG Pipelines with Fine Grained Authorization
33mHow to Prevent AI Agents from Accessing Unauthorized Data
29mHow to Prevent AI Agents from Accessing Unauthorized Data
29mHow Google Built a Consistent, Global Authorization System with Zanzibar
47mFine Grained Authorisation with Relationship Based Access Control
46mRebuilding Access Management with ReBAC