A Brief Guide to Secure Agentic Knowledge Retrieval
summary
Explains how OpenSearch indexes enforce source permissions for agentic RAG by using admin connectors instead of per‑user tokens, storing an ACL field with tokens per document plus a principal key‑value store to build query‑time filters, and normalizing principals (user principal name). Claims 12‑hour index freshness is typical, but high‑scale use (example: central bank changing 2 million permissions nightly) requires event‑driven updates or last‑mile checks; demo shows permissions prevent unwanted disclosure.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:01Agentic workflows and security
- 3:07Identity access management
- 4:53OpenSearch retrieval
- 5:34User vs admin connectors
- 7:17Permission-based indexing
- 8:23ACLs and principal store
- 10:29User-group relationships
- 12:14Demo: permission-based search
- 13:41Connector spread risks
- 14:22Q&A: ACL changes
- 17:51Edge cases and indexing
- 19:07Agent vs human permissions
- 20:10Q&A: connector challenges
- 21:05Closing
related talks
17mFine-Grained Authorization for Secure RAG With OpenSearch
30mFine-Grained Authorization: The Missing Piece in Agentic AI Security
57mFine-Grained Authorization: The Missing Piece in Agentic AI Security
23mZero-Trust Architecture for Agentic AI: Securing Multi-User Access and Third-Party Integrations
30mFuture Paths for the OpenSearch Security Layer
30mYour AI Agents Are Overprivileged: The Case for Fine-Grained Authorization
33mHow to Prevent AI Agents from Accessing Unauthorized Data
19mPersona-based access to enterprise data for generative AI apps