Identity for AI Agents

AI agents are becoming digital workers. Digital workers need identities.

As autonomous agents begin interacting with enterprise systems, a fundamental security question emerges:

Who is the agent?

Allowing multiple agents to operate through one shared administrator account creates significant security and governance problems.

My research explores treating AI agents as first-class non-human identities.

An agent could have its own:

• Identity
• Role
• Permissions
• Authentication
• Authorization policies
• Temporary credentials
• Audit history
• Lifecycle
• Resource boundaries

Permissions could be granted just in time and automatically revoked when the task is completed.

This brings established Identity and Access Management principles into the emerging world of agentic AI.