Science & Tech · AI · 6 hrs ago
Nvidia unveils OpenShell architecture to govern AI agents
Nvidia introduced a plan for governing AI agents called the OpenShell Reference Architecture.
It is designed for companies that use AI agents to carry out tasks, while keeping their actions within defined limits.
The system builds on Nvidia’s OpenShell runtime to manage agents, monitor their behavior and control access to credentials and tools.
Nvidia’s blueprint requires its hardware, including Vera Rubin infrastructure and BlueField-4 data-processing units.
Technology companies and cloud providers are among the partners, and Palantir is working with Nvidia and customers on deployments.
Banks BBVA and Santander are reviewing the design, while SAP is integrating it with its Joule Studio runtime.
OpenShell collects telemetry by default, though users can opt out, and some security issues remain listed on GitHub; no deployment timetable was given.
Nvidia introduced its OpenShell Reference Architecture to help organizations create, operate, monitor and govern AI agents.
The architecture is based on Nvidia’s OpenShell secure runtime and is designed to set boundaries around agents’ access and behavior.
Nvidia’s blueprint uses its Vera Rubin infrastructure, Vera CPUs and BlueField-4 DPUs, so organizations need Nvidia hardware to implement it.
Nvidia says the architecture is available to partners, and names technology companies and cloud providers among those involved.
BBVA and Santander are reviewing the specification, while SAP is integrating it with Joule Studio; Palantir is also working with Nvidia and customers on deployments.
- Who
- Nvidia introduced the architecture. Palantir is also working with Nvidia and customers on deployments.
- What
- Nvidia unveiled its OpenShell Reference Architecture for operating and governing AI agents.
- When
- Nvidia introduced it on Wednesday, October 21, 2026.
- Where
- Not stated.
- Why
- To help keep AI agents’ activities within defined boundaries and improve their monitoring and governance.
This story does not have two clearly opposing sides.
This story does not have a timeline yet.
- Architecture
- OpenShell Reference Architecture
- Runtime
- Nvidia OpenShell secure runtime
- Hardware
- Vera Rubin infrastructure, Vera CPUs and BlueField-4 DPUs
- Reviewing the specification
- BBVA and Santander
- Integration
- SAP is integrating the architecture with its Joule Studio runtime
- Deployment work
- Palantir is working with Nvidia and customers
4 sources
What people are saying
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.
Third-party services, model providers, inference endpoints, agents, or tools that you configure and use with OpenShell may have their own terms and privacy practices.