Agentic Identity

Confirming that an AI agent acting on someone's behalf is tied to a real, verified person and staying within what that person actually authorized it to do.

For example, if a person lets an AI assistant book travel for them, agentic identity is what links the assistant to that person and defines how much it can spend.

A verification layer that links an AI agent to a verified human owner, defines the agent's permitted scope of action, and continuously monitors its behavior for misuse, rather than verifying the agent once at setup and never again.

Agentic identity has three working parts:

• Ownership, which ties the agent to a verified person or organization.
• Scope, which defines what the agent may do, through a mandate that sets actions and limits.
• Monitoring, which watches the agent's behavior and flags activity outside its scope.

Agentic identity sits next to terms that are easy to mix up:

• Know Your Agent (KYA) is the practice of verifying an agent and the people behind it, and agentic identity is the technical layer that supports it.
• Agent mandate is the specific permission a person gives an agent, which is one part of agentic identity.
• Authentication confirms a person at login, while agentic identity must also confirm what an agent is allowed to do afterward.

For more, see What the IMF Got Right About Know Your Agent and The Agentic Fraud Report.