Agents Are Not Users or Services: Securing AI Traffic with a New Identity Model
summary
Claims that agents require a distinct identity model: introduce an actor claim (OAuth 2.1 draft) to record user→agent→subagent delegation and preserve provenance; show real failure patterns—over‑permissioned tool access, confused‑deputy flows, cross‑user data leaks and missing audit trails—and quantify that per‑hop identity is required. Demonstration uses WSO2 AI Gateway (open source) to proxy LLMs and MCPs, vault provider keys, issue bounded tokens, enforce policies, rate/cost limits, prompt/PII guardrails and capture observability for production MCP/A2A agent architectures.
AI-written summary · about
chapters (12)
- 0:00Agent identity challenges
- 1:11Agenda
- 2:36Agent scenario walkthrough
- 4:43Agent identity gap
- 7:54Securing multi-component agents
- 10:28Risks of JWT tokens
- 12:38Agent authorization questions
- 14:03OAuth 2.1 delegation
- 16:10AI gateways role
- 18:38WSO2 AI Gateway
- 22:55Agent security checklist
- 23:16Conclusion
related talks
20mCIAM for AI: Authn/Authz for Agents
17mDon't give your agent the master key
40mIn AI We Trust? Securing the Future, One Agent at a Time
04mYou Don't Need a New Identity Platform for the AI Era
23mWho is Calling? Identity for Agents with AAuth
23mZero-Trust Architecture for Agentic AI: Securing Multi-User Access and Third-Party Integrations
22mAccountable AI agents
25mWho Let the Agents Out? Securing AI Workflows the Right Way