OpenFGA For Agents: Safe Delegation In 5 Minutes
summary
Shows a concrete OpenFGA relationship model for delegation (human → agent → tool → resource) and a real check flow that enforces ephemeral per-task permissions. Agents start with zero privileges; an intent LM infers needed scopes, then OpenFGA issues least-privilege, auditable, revocable tuples. Demo uses Linear and Slack and prevents a prompt-injection that requests extra permissions by denying the unauthorized email send; two modes run: interactive approval and autonomous cancellation when inferred permissions don’t match.
AI-written summary · about
related talks
57mFine-Grained Authorization: The Missing Piece in Agentic AI Security
54mFine-Grained Authorization for Modern Applications
37mRole-Based Access Is so Yesterday: Revolutionizing Authorization with Open FGA
30mFine-Grained Authorization: The Missing Piece in Agentic AI Security
33mFine-Grained Policies RBAC with OpenFGA
25mImplementing Modern Cloud Native Authorization Using OpenFGA
06mProject Lightning Talk: OpenFGA: The Cloud Native Way to Implement Fine Grained Authorization
55mYour AI Agents Are Overprivileged: The Case for Fine-Grained Authorization