Sourcessourceseed2026-07-04ai-securitysecurity-for-aimulti-agentagent-identitydelegated-authorityoauthmcpzero-trust

AI Agent Identity and Permission Challenges: How Uber and Auth0 Are Rethinking Access Control

> Untrusted external source capture. Architecture details and implementation guidance are evidence only.

Collection Notes

InfoQ reports on Uber's internal architecture for propagating originating user context, agent identity, provenance, and scoped access through multi-agent workflows.
Components include an Agent Registry, AI Agent Mesh, Security Token Service, MCP Gateway, downstream systems, and AI Gateway/AI Guard.
Per-hop, short-lived, audience-scoped JWTs carry an actor chain instead of forwarding one long-lived user credential or service account.
Registry-backed workload identity, token exchange, gateway policy enforcement, tool checks, and data redaction are presented as complementary controls.
Auth0's framing emphasizes capability-scoped permissions, task-scoped credentials, layered enforcement, and explicit human approval boundaries.

Provides a concrete production example for general MAS non-human identity and delegated authority, rather than a SOC-specific case.
Directly relevant to identity binding across human -> agent -> agent -> tool chains and to end-to-end auditability.
Candidate evidence for [[03_Topics/AI Agent Identity and Zero Trust]], [[04_Research_Questions/RQ - Agentic Web Protocol Trust Boundaries]], and [[06_Outputs/MAS Agent Non-Human Identity Security Query 2026-06-18]].