NVIDIA’s OpenShell Reference Architecture is designed to help businesses control AI agents operating across enterprise infrastructure and local computing systems.
It builds on NVIDIA’s existing OpenShell security software, combining it with hardware for running AI models, managing agent operations and enforcing security rules. The aim is to let agents carry out useful work within defined boundaries, while giving businesses tools to monitor their behaviour and stop activity that crosses those limits.
“AI capability and safety must advance together,” said Jensen Huang, CEO of NVIDIA. “As enterprises prepare to deploy billions of AI agents, safety and control must be built into the computing architecture. NVIDIA created OpenShell to give agents the freedom to do useful work within clearly defined boundaries — contained, monitored and under control. With the OpenShell Reference Architecture, agents can think on Rubin, act on Vera and be governed on BlueField.”
How the system works
The architecture divides the work between three main parts of NVIDIA’s computing platform.
Vera Rubin systems run AI models. Vera CPUs handle the software that manages agents and their use of tools, while running OpenShell to provide a restricted environment for agent operations.
BlueField-4 data processing units run NVIDIA Sentry security software. This provides a separate monitoring and enforcement layer, checking agent behaviour and enforcing security policies through the hardware.
NVIDIA says that if an agent attempts to move outside its software boundary, Sentry can isolate and stop it within milliseconds. The company also says Sentry running on BlueField-4 can increase agent capacity by up to 20%.
The architecture extends beyond large computing installations to support NVIDIA RTX Spark, DGX Spark and DGX Station systems for agents operating locally across industries including retail, healthcare, manufacturing, robotics and telecommunications.
Partners work on business applications and cloud support
NVIDIA’s partners are working on several parts of the system, from supporting cloud deployments to connecting agents with business software.
Enterprise software providers are building agent platforms and integrating OpenShell with tools that govern business operations. These integrations connect company policies with controls enforced while an agent works.
ServiceNow’s AI Passport integration allows businesses to revoke an agent’s access across software and hardware within seconds, according to NVIDIA. Other partners are working to deploy secure agents in production and incorporate OpenShell into model development.
The architecture supports Linux and Kubernetes infrastructure, as well as Microsoft Windows through Microsoft Execution Containers. It can run on cloud operating systems from Amazon Web Services, Google Cloud and Microsoft Azure.
Security tools and supporting infrastructure
Cybersecurity providers are integrating support for the architecture into their agent security products. Hardware manufacturers and cloud providers are building the infrastructure needed to run it.
Banks, including BBVA and Santander, are exploring the OpenShell Reference Architecture as a framework for strengthening agent safety, security and oversight.
The OpenShell Reference Architecture is available now, NVIDIA said.
