AI Agents Need Their Own Identity
September 20, 2026 · Keith Pasley
We have spent years building identity systems around two basic ideas: people and applications.
AI agents don’t fit neatly into either one.
An agent can act on behalf of a person, call another agent, connect to an API, access an MCP server and execute actions across multiple systems. It may also operate for hours without someone sitting behind a keyboard.
That creates a different identity problem.
This week, WSO2 made its Agent Manager generally available with per-agent identity, role-based access, delegation, token exchange and instant revocation, along with governance at the agent, MCP and model levels.
I think the bigger story is what this says about enterprise architecture.
We are starting to need an identity layer designed specifically for non-human actors that can make decisions and take actions.
Knowing which employee launched an agent isn’t enough.
We also need to know:
Which agent is this?
Who or what is it acting for?
What has it been authorized to do?
Which tools and data can it reach?
Can we revoke that authority immediately?
This becomes even more important when agents start interacting with other agents.
Identity then becomes part of the control plane for agentic AI, not simply another login mechanism.
We already manage identities for people, devices, workloads and applications.
Agents may be the next identity class enterprises have to manage.
Technology doesn’t create business value. Better decisions do.