I’ve worked with several security teams who were confident they knew how many agents were running in their environment. That confidence didn’t survive the assessment. What we uncovered were hundreds of agents operating on inherited authority, cached credentials, shared service accounts and user sessions that were never designed to be reused by autonomous systems. No one had authorized them. No one owned them. And no one knew the full scope of what they could do.

The cybersecurity industry has spent the last year asking the wrong question about AI agents. Most have been focused on what agents can access, what permissions they hold and how to revoke them. Those are valid concerns, but they only scratch the surface. The deeper question is about delegation: who authorized this agent to act, on whose behalf, under what conditions and what happens when the chain of trust breaks.

44% of organizations rank non-human identity and AI agent security as their top IAM priority over the next 12 to 24 months. But priority and readiness are not the same thing. 

No single vendor is covering every part of the governance gap. The market is maturing quickly, but the best choice today is the one that closes the highest-priority gaps and can be operationally sustained by the team. Three steps to start:

Not a one-time discovery sweep, but a registry. Every agent in the environment needs to be identified, bound to a named owner and assessed against a defined purpose.

Knowing who the agent is, its sponsor, its purpose and what it is allowed to touch can start now. How each agent proves its identity cryptographically at runtime, where emerging standards for workload identity such as SPIFFE and SPIRE are still maturing, is a decision that can wait for the market to settle.

Kevin Converse is the Vice President, Identity and Access Management at GuidePoint Security. He has over 20 years of experience in the IT and cybersecurity domains. In his career, Kevin spent several years working for a large financial organization and higher education institution where he oversaw projects ranging from end user deployment, remote workforce enablement and infrastructure automation. He has a decade’s worth of experience across multiple cybersecurity domains including vulnerability management, SIEM integration and security architecture and identity management. He has spent the last 5 in the consulting space deploying identity and access management programs across multiple verticals.