The Agentic Enterprise Needs a New Security Architecture
Cyera, Tuesday, November 10th, 2026
Cyera argues that securing AI agents requires an intelligent control plane that coordinates existing controls, not more point products.
A Cyera executive writes that enterprises are adding endpoint, network, gateway and API controls for AI agents faster than they are building the intelligence that tells those controls what to do.
No single control can decide whether an agent should take a given action on given data for a given business purpose, as a procurement agent that posts negotiated prices to an external service illustrates.
The post says agents break five assumptions behind enterprise security, including human speed, dormant access and predictable application behavior.
It cites research by Cyera and Oso covering 2.4 million workers and 3.6 billion permissions, which found people used only about 4% of the access they had been granted.