#Agentic AI
Governing Autonomous Systems: Auditability, Non-Repudiation, and Revocation
Getting identity, delegation, and consent right in the moment is not enough if you cannot prove what happened afterward. Learn how token introspection (RFC 7662), revocation (RFC 7009), and captured decision chains turn governance into a real, provable control.
Least Privilege at the Data Layer: Relationship-Based Authorization for RAG
Scopes and roles answer whether an agent can call an API, not whether it can retrieve this specific record for this specific user right now. Learn why ReBAC, the model behind Google Zanzibar, is what agentic RAG pipelines actually need.
Securing the New Attack Surface: Authorization for Tool-Calling Agents and MCP
Handing an agent a tool means standing up a new API client, with every authentication and authorization obligation that implies. Learn how OAuth 2.1, dynamic client registration, and resource indicators secure MCP and tool-calling agents properly.
Beyond the Consent Screen: Agent-to-Agent and Agent-to-App Trust
Per-application OAuth consent doesn't scale once agents reach into other apps and other organizations' agents. Learn why letting the identity provider mediate agent-to-agent trust — the way it already mediates SSO — is the pattern that scales.
Human-in-the-Loop, Automated: Asynchronous Authorization for High-Stakes Agent Actions
A one-time consent checkbox for AI agents is meaningless, and synchronous approval defeats the point of autonomy. Learn how CIBA and Rich Authorization Requests (RFC 9396) combine into real, asynchronous, non-repudiable human-in-the-loop approval.
Delegation Done Right: Token Exchange and the On-Behalf-Of Pattern
Storing user credentials, forwarding tokens verbatim, or forcing constant re-authentication are the three ways teams get agent delegation wrong. Learn how OAuth 2.0 Token Exchange (RFC 8693) and JWT Bearer assertions (RFC 7523) solve it correctly.
Agent Identity: Why Your AI Agent Needs to Be a First-Class Principal
Giving an AI agent the user's token or a shared service-account credential collapses least privilege and attribution. Learn why an agent needs to be a distinct, first-class principal — and how RFC 8693's act claim already models it.
The Authentication and Authorization Challenges of Agentic AI
Agentic AI breaks the two-actor identity model everything else was built on. This pillar post maps the seven authentication and authorization gaps — identity, delegation, consent, trust chains, tool surfaces, data scope, and governance — that the rest of the series covers one at a time.