Nutanix Proposes Three-Layer Security Architecture for AI Agents
Nutanix has introduced a three-layer, defense-in-depth security architecture to address the risks posed by autonomous AI agents. The new model aims to manage threats that traditional application-level controls cannot contain.
Nutanix has outlined a new three-layer security architecture designed to tackle the growing risks associated with autonomous AI agents. The company argues that conventional application-level security measures are insufficient to manage agents capable of independent decision-making and action.
Oscar Wahlberg, senior director of product management at Nutanix, highlighted that these agents can introduce risks such as accidental database deletion or sensitive data leakage. This problem escalates as agents move from experimental phases into production environments. To counter this, Nutanix advocates for a defense-in-depth strategy spanning infrastructure, storage, compute, networking, and a governing control plane.
The foundational layer, infrastructure, focuses on establishing trust in the environment where AI agents operate. This includes hardware-rooted trust mechanisms like platform attestation and secure boot, alongside access controls. The network layer governs agent-to-agent and agent-to-system communications through dynamic policies, preventing unauthorized access and data exfiltration.
The third layer, the control plane, serves as a centralized management point. It enables oversight of agent permissions, tool access, and resource consumption, ensuring consistent policy enforcement. This layer is designed to mitigate risks such as privilege misuse and runaway agents.
Wahlberg stressed that a one-size-fits-all security approach is inadequate for agentic AI environments. Nutanix's solution, Agent Gateway, aims to provide manageability and cost control for autonomous agent deployments, with integrations including those from Cisco Systems.