How to Secure Agents using OAuth
summary
Argues that long‑lived API keys must replace API keys with OAuth and shows concrete fixes: MCP’s first draft ships with no auth, the followup spec separates an authorization server from resource servers after ~400 PR comments, and current best practices use client_credentials for agent-to-agent calls, token exchange and the identity assertion grant for cross‑domain chaining. Calls out failures of dynamic client registration (a 10‑year spec with little adoption) and proposes URL+PKI for agent identity, attestation, and rich authorization requests; Keycard implements these standards.
AI-written summary · about
chapters (13)
- 0:00Introduction and speaker intro
- 1:06Security risks of static secrets
- 3:11OAuth basics and flows
- 6:37Applying OAuth to MCP
- 7:40MCP authorization spec issues
- 10:09Fixed OAuth model in MCP
- 10:50Future OAuth needs for AI
- 13:37Agent identity solutions
- 14:39Agent attestation and LLMs
- 15:43Transactional authorization
- 16:24Chain of custody in APIs
- 17:26Async interactions and real-time access
- 18:09Future of AI security and closing
related talks
52mAI Agents Need Permission Slips
01h 22mIdentity for AI Agents
46mA Deep Dive into the Past, Present & Future of OAuth
25mIdentities and Authentication for Your Agents With Keycloak
23mIdentity Best Practices to Secure & Scale Agentic AI / MCP Deployments
27mDesigning an Agentic AI-Ready Secure Authentication and Authorization Platform
32mAI Agents with Keycloak in MCP
56mSecuring an API Ecosystem with OAuth 2 0