---
title: 'Nvidia launches the OpenShell Reference Architecture for governing AI agents'
url: 'https://securteksecurity.com/news/nvidia-openshell-reference-architecture-agent-governance'
markdown: 'https://securteksecurity.com/news/nvidia-openshell-reference-architecture-agent-governance.md'
date: '2026-10-21'
description: "Sourcing note: this report rests on a single reporting chain. Nvidia's own launch release, distributed by GlobeNewswire at GTC Berlin on October 21, 2026, is the source for the architecture's specifications, performance figures and partner list, and the other coverage found, including Unite.AI's re…"
taxonomy:
  category:
    - AI
---

# [Nvidia launches the OpenShell Reference Architecture for governing AI agents](https://securteksecurity.com/news/nvidia-openshell-reference-architecture-agent-governance)

   21st Oct 2026  

Sourcing note: this report rests on a single reporting chain. Nvidia's own launch release, distributed by GlobeNewswire at GTC Berlin on October 21, 2026, is the source for the architecture's specifications, performance figures and partner list, and the other coverage found, including Unite.AI's report, derives from that release and Nvidia's technical posts. The performance figures are Nvidia's own claims and have not been independently benchmarked. This item will be updated when independent coverage or customer deployments appear.

At GTC Berlin on October 21, 2026, Nvidia announced the OpenShell Reference Architecture, a full-stack design for building, containing, observing and enforcing safe AI agents across enterprise infrastructure and edge systems, available now. It extends Nvidia's open-source OpenShell secure runtime: models run on Vera Rubin infrastructure, agent harnesses act on Vera CPUs, and BlueField-4 DPUs run Nvidia Sentry, which provides continuous out-of-band monitoring of agent behavior and in-silicon policy enforcement, quarantining an agent that moves outside its boundary in milliseconds, according to Nvidia. The design extends to agents at the edge on RTX Spark, DGX Spark and DGX Station systems. Nvidia lists Anthropic, AWS, Cisco, CrowdStrike, Google Cloud, Meta, Microsoft, OpenAI, Palo Alto Networks, SAP, Siemens and SpaceX Super Intelligence among those working with the architecture, with BBVA and Santander exploring it for agent governance.

Sources: [GlobeNewswire](https://www.globenewswire.com/news-release/2026/10/21/3385311/0/en/nvidia-launches-openshell-reference-architecture-for-safe-ai-agents.html); [Unite.AI](https://www.unite.ai/nvidia-debuts-openshell-reference-architecture-for-agent-governance/)

---

## Navigation

- Parent: [News](https://securteksecurity.com/news.md)
- Previous: [Nvidia invests $3 billion in Meta and deepens Muse partnership as offering rises to $97.5 billion](https://securteksecurity.com/news/nvidia-invests-3-billion-meta-muse-partnership-97-5-billion-offering.md)
- Next: [NYSE American Options trading halted on a technical issue](https://securteksecurity.com/news/nyse-american-options-trading-halt-technical-issue.md)