AI agent IAM: why enterprise identity governance is broken

Enterprise agentic AI is stalled not by model quality, but by the inability to govern non-human identities at scale.

Cisco’s Jeetu Patel said about 85% of AI deployments remain in pilot, while only 5% reach production.

Michael Dickman argued security and policy enforcement must be built in from day one, with network visibility helping control agent behavior.

The core issue is an architecture and trust problem: without mature access controls, agents can expand risk, expose sensitive systems, and slow adoption.