NVIDIA Launches OpenShell Reference Architecture to Strengthen AI Agent Operational Security and Hardware-Level Controls

Odaily reports: NVIDIA has launched the NVIDIA OpenShell reference architecture, designed for building, running, monitoring, and constraining AI Agents across enterprise infrastructure and edge systems, while strengthening governance, isolation, and security controls. The architecture is built on the OpenShell secure runtime and combines NVIDIA Vera Rubin, Vera CPU, and BlueField-4 DPU.

Among these, Vera Rubin handles model inference, Vera CPU is used to run the Agent harness and tool calls, and BlueField-4 continuously monitors Agent behavior and enforces hardware-level policy controls through NVIDIA Sentry; when an Agent exceeds established software boundaries, Sentry can isolate and halt it within milliseconds.