19. Agent Identity Architect
Evidence label: Capability. Agent identity is a distinct control problem, but current postings usually assign it to AI Security, IAM, platform security, or AI Platform roles rather than an Agent Identity Architect family.
What they do
Design and enforce the identity layer for AI agents: how agents authenticate to services, what permissions they hold, how those permissions are scoped and audited, and how privilege escalation is prevented when agents chain tool calls across services.
Why the capability matters
Agents combine delegated authority, access to private data, exposure to untrusted content, and external actions. Shared credentials or inherited permissions make attribution and least privilege difficult. The control boundary therefore needs an identity per workload or session, scoped authorization, auditable delegation, and explicit handling of sub-agent privileges.
What the role covers
- Workload identity: issue identities per agent workload or session instead of sharing team credentials.
- Delegated authorization: bind each action to a user, service, task, and allowed scope with expiry and revocation.
- Tool enforcement: validate tool calls against the current task scope at an identity-aware gateway or equivalent policy point.
- Session tracing: attribute each action to a specific agent session and delegation chain, not just "the AI system."
- Sub-agent controls: define permission inheritance explicitly and reduce child scope to the minimum required for the delegated task.
Required skills
IAM and OAuth/OIDC expertise, zero-trust architecture, Kubernetes RBAC, understanding of MCP security model, incident response for non-deterministic systems.
Entry paths
Cloud security engineer, identity/access management specialist, platform engineer with security focus.