Your AI Agents Are Overprivileged - The Case for Fine-Grained Authorization

Explore the critical security challenges of AI agent authorization in this 30-minute conference talk from the Linux Foundation's Open Source Summit. Learn why traditional access control systems fail to handle the contextual, document-level permissions required for modern AI applications at enterprise scale. Discover how Fine-Grained Authorization (FGA) provides essential security for Retrieval-Augmented Generation (RAG) and agentic AI systems by ensuring AI agents respect strict data boundaries. Examine real-world implementations using OpenFGA and LangChain to address multi-tenant isolation, prevent data leakage, and maintain comprehensive auditability while scaling to billions of access decisions without performance degradation. Gain practical insights for securing both internal AI copilots and customer-facing AI applications as generative AI adoption continues to accelerate across enterprises.