Nvidia proposes OpenShell architecture to govern AI agents
Original title: Nvidia picks OpenShell to tame AI gents
The blueprint needs Vera Rubin, Vera CPUs and BlueField-4, while BBVA and Santander are only reviewing it.
Nvidia introduced an OpenShell Reference Architecture for creating, operating, observing, and policing AI agents on its hardware. Models such as Nemotron would run on Vera Rubin, harnesses on Vera CPUs, and BlueField-4 DPUs would run Sentry to watch behavior. Linux distributors and major clouds are named partners, and SAP is integrating the design with Joule Studio. OpenShell still lists open security issues on GitHub and collects telemetry by default unless users opt out.
Unconfirmed: BBVA and Santander are described as reviewing the architecture, not deploying it.
- Huang said agents can think on Rubin, act on Vera, and be governed on BlueField.
- Named model builders include Anthropic, OpenAI, Google, Meta, and others.
- BBVA and Santander are reviewing the spec against governance requirements.
- Telemetry is collected by default, with an opt-out available.
From the original article
2 min read at theregister.comGive your agents freedom to do useful work by locking them down
By Thomas Claburn · theregister.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.