Why Traditional IAM Falls Short for AI Agents

Summary
The article argues that human-centric IAM cannot adequately govern AI agents across API workflows. It recommends application-centric runtime authorization with short-lived, least-privilege tokens using OAuth 2.0, token exchange and dynamic client registration.
Key points
- The article says non-human identities account for more than 90% of authentications and that AI agents can trigger chains of API calls.
- It argues that treating short-lived agents as persistent directory identities can create identity sprawl and orphaned permissions.
- The proposed model evaluates authorization continuously, based on the application, action, data and context.
- It recommends short-lived, just-in-time, least-privilege tokens that carry information about the acting agent, represented user and requested action.
- The article identifies OAuth 2.0, token exchange and dynamic client registration as existing standards that can support this model.
- Curity Access Intelligence is presented as a product intended to address the runtime-governance gap.
Article Details
- Topic
- Runtime authorization and application-centric identity and access management for AI agents and machine identities
