Autonomous AI agents that reason and act across enterprise systems introduce risks traditional security controls can’t contain, according to Nutanix. A single-layer approach leaves gaps that allow agents to misuse credentials, leak data, or delete databases despite passing initial filters, said Oscar Wahlberg, senior director of product management at Nutanix.
Nutanix outlines a defense-in-depth architecture with three distinct layers: infrastructure, network, and control plane. The infrastructure layer establishes hardware-rooted trust to verify agent identities and prevent tampering, while the network layer governs dynamic agent communications to block lateral movement and data exfiltration. The control plane centralizes policy enforcement, token governance, and runtime visibility to prevent privilege misuse and runaway agents.
The company highlights a partnership with Intel and Cisco to deliver a turnkey stack. Intel provides hardware-rooted execution environments, Cisco supplies secure networking fabrics, and Nutanix unifies governance across the stack. Wahlberg warns that enterprises often underestimate the control plane’s role in scaling secure agent deployments, emphasizing its need for continuous observability and token budgeting in production.
