# NVIDIA announces OpenShell Reference Architecture for AI agents [Canonical reading](https://escapemeridian.com/readings/nvidia-announces-openshell-reference-architecture-for-ai-agents-9ec7aa46) NVIDIA says its OpenShell Reference Architecture combines hardware and software to build, contain, observe and enforce rules for AI agents across enterprise infrastructure and edge systems. The company describes it as an extension of its OpenShell secure runtime software, with NVIDIA Vera Rubin infrastructure for model workloads, NVIDIA Vera CPUs for agent harnesses and tool use, and NVIDIA BlueField-4 DPUs running NVIDIA Sentry for monitoring and enforcement. NVIDIA says NVIDIA Sentry can quarantine and stop an agent that attempts to leave its software boundary. The company also claims performance benefits, including higher throughput per megawatt for agentic workloads and increased agent capacity with NVIDIA Sentry; these are NVIDIA’s stated figures, not independent evidence of real-world reliability or safety. The announcement names operating-system, cloud, model, cybersecurity and enterprise software partners working with the architecture. It also describes some deployments as explorations: BBVA and Santander are working with NVIDIA to explore it, while NVIDIA says Palantir and NVIDIA are working with customers to deploy secure agents in production. The announcement says the architecture is available, but does not report results from deployments or explain how its enforcement performs across different workloads. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source NVIDIA · Article Original source date: Oct 21, 2026 (2026-10-21T09:48:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-05T17:31:23.097Z. First observed: 2026-10-21T09:57:27.493Z. Retrieved: 2026-10-22T23:01:49.245Z. - [NVIDIA Launches OpenShell Reference Architecture for Safe AI Agents](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Limits The performance and security statements are NVIDIA’s claims; the announcement provides no independent evaluation or deployment results. The announcement describes some partner activity as exploration, and does not establish that all named partners have deployed the architecture. The announcement does not explain how enforcement performs across different workloads or how its software boundaries are defined. ## Open question How does NVIDIA Sentry’s enforcement perform in real deployments, including when agents encounter workloads or actions not covered by their software boundaries? ## Source-stated NVIDIA announced an architecture combining its OpenShell software with NVIDIA compute and security components for agent development and operations. > The OpenShell Reference Architecture builds on the foundation of NVIDIA OpenShell™ secure runtime software, extending it into full-stack, industrial-scale infrastructure for agent development and operations. - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source-stated NVIDIA says BlueField and NVIDIA Sentry continuously monitor agent behavior and enforce policies in silicon; NVIDIA Sentry can quarantine and stop agents that cross their software boundary. > if an agent attempts to move outside its software boundary, NVIDIA Sentry quarantines and stops it in milliseconds. - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source-stated NVIDIA claims its Vera Rubin NVL72 systems deliver higher throughput per megawatt for agentic workloads, and that NVIDIA Sentry boosts agent capacity. > NVIDIA Vera Rubin NVL72 systems deliver 30x higher throughput per megawatt. - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source-stated The announcement says BBVA and Santander are exploring the architecture with NVIDIA, while Palantir and NVIDIA are working with customers to deploy secure agents in production. > Leading banks — including Europe’s BBVA and Santander — are working with NVIDIA to explore the OpenShell Reference Architecture - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source-stated NVIDIA says the OpenShell Reference Architecture is available. > The NVIDIA OpenShell Reference Architecture is now available. - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source and contributor context NVIDIA · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media declared by the original publisher; availability checked separately from the editorial reading. - [NVIDIA OpenShell Reference Architecture · NVIDIA](https://iprsoftwaremedia.com/219/files/202610/b61bfda00ad668f9890d5b5b0d775b1e/6ad693aa3d63324483072858_NVIDIA-OpenShell-Reference-Architecture/NVIDIA-OpenShell-Reference-Architecture_21af43b3-e9a2-4b8b-a792-e5382e40f9aa-prv.jpg?v=21af43b3-e9a2-4b8b-a792-e5382e40f9aa) ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## CPU The central processing unit: a general-purpose processor that carries out a computer's instructions. - [Definition reference](https://aws.amazon.com/what-is/cpu/) ## Tool use A model's ability to operate other software, such as a search engine, a calculator or a code runner, while it works. - [Definition reference](https://docs.anthropic.com/en/docs/build-with-claude/tool-use/overview) ## Megawatt A unit of power equal to a million watts. - [Definition reference](https://en.wikipedia.org/wiki/Watt) ## Agentic AI AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer. - [Definition reference](https://developers.google.com/machine-learning/glossary#agentic) ## NVIDIA NVIDIA develops processors, networking and software used for AI, graphics and scientific computing. - [Explore this context](/entities/nvidia) - [Reference](https://blogs.nvidia.com/) ## Editorial provenance Version 1. Publication ID: 268e09db-5934-4a7e-8f78-40da768374d7. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-05T17:31:23.097Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 874bad201578a9b521bf37a4ab60d60e6b9729615079b84eca2a97f35711798f. Source text hash: b10d79fff8600e8faca4ae0010cc767e3c4fff3e610fddd44c16073f4b6f19a7. ## Revision and source-check history ### Version 1 — 2026-11-05T17:31:23.097Z NVIDIA says its OpenShell Reference Architecture combines hardware and software to build, contain, observe and enforce rules for AI agents across enterprise infrastructure and edge systems. The company describes it as an extension of its OpenShell secure runtime software, with NVIDIA Vera Rubin infrastructure for model workloads, NVIDIA Vera CPUs for agent harnesses and tool use, and NVIDIA BlueField-4 DPUs running NVIDIA Sentry for monitoring and enforcement. NVIDIA says NVIDIA Sentry can quarantine and stop an agent that attempts to leave its software boundary. The company also claims performance benefits, including higher throughput per megawatt for agentic workloads and increased agent capacity with NVIDIA Sentry; these are NVIDIA’s stated figures, not independent evidence of real-world reliability or safety. The announcement names operating-system, cloud, model, cybersecurity and enterprise software partners working with the architecture. It also describes some deployments as explorations: BBVA and Santander are working with NVIDIA to explore it, while NVIDIA says Palantir and NVIDIA are working with customers to deploy secure agents in production. The announcement says the architecture is available, but does not report results from deployments or explain how its enforcement performs across different workloads. ## Original source NVIDIA · Article Original source date: Oct 21, 2026 (2026-10-21T09:48:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-05T17:31:23.097Z. First observed: 2026-10-21T09:57:27.493Z. Retrieved: 2026-10-22T23:01:49.245Z. - [NVIDIA Launches OpenShell Reference Architecture for Safe AI Agents](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Limits The performance and security statements are NVIDIA’s claims; the announcement provides no independent evaluation or deployment results. The announcement describes some partner activity as exploration, and does not establish that all named partners have deployed the architecture. The announcement does not explain how enforcement performs across different workloads or how its software boundaries are defined. ## Open question How does NVIDIA Sentry’s enforcement perform in real deployments, including when agents encounter workloads or actions not covered by their software boundaries? ## Source-stated NVIDIA announced an architecture combining its OpenShell software with NVIDIA compute and security components for agent development and operations. > The OpenShell Reference Architecture builds on the foundation of NVIDIA OpenShell™ secure runtime software, extending it into full-stack, industrial-scale infrastructure for agent development and operations. - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source-stated NVIDIA says BlueField and NVIDIA Sentry continuously monitor agent behavior and enforce policies in silicon; NVIDIA Sentry can quarantine and stop agents that cross their software boundary. > if an agent attempts to move outside its software boundary, NVIDIA Sentry quarantines and stops it in milliseconds. - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source-stated NVIDIA claims its Vera Rubin NVL72 systems deliver higher throughput per megawatt for agentic workloads, and that NVIDIA Sentry boosts agent capacity. > NVIDIA Vera Rubin NVL72 systems deliver 30x higher throughput per megawatt. - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source-stated The announcement says BBVA and Santander are exploring the architecture with NVIDIA, while Palantir and NVIDIA are working with customers to deploy secure agents in production. > Leading banks — including Europe’s BBVA and Santander — are working with NVIDIA to explore the OpenShell Reference Architecture - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source-stated NVIDIA says the OpenShell Reference Architecture is available. > The NVIDIA OpenShell Reference Architecture is now available. - [Supporting source](https://nvidianews.nvidia.com/news/openshell-reference-architecture) ## Source and contributor context NVIDIA · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media declared by the original publisher; availability checked separately from the editorial reading. - [NVIDIA OpenShell Reference Architecture · NVIDIA](https://iprsoftwaremedia.com/219/files/202610/b61bfda00ad668f9890d5b5b0d775b1e/6ad693aa3d63324483072858_NVIDIA-OpenShell-Reference-Architecture/NVIDIA-OpenShell-Reference-Architecture_21af43b3-e9a2-4b8b-a792-e5382e40f9aa-prv.jpg?v=21af43b3-e9a2-4b8b-a792-e5382e40f9aa) ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## CPU The central processing unit: a general-purpose processor that carries out a computer's instructions. - [Definition reference](https://aws.amazon.com/what-is/cpu/) ## Tool use A model's ability to operate other software, such as a search engine, a calculator or a code runner, while it works. - [Definition reference](https://docs.anthropic.com/en/docs/build-with-claude/tool-use/overview) ## Megawatt A unit of power equal to a million watts. - [Definition reference](https://en.wikipedia.org/wiki/Watt) ## Agentic AI AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer. - [Definition reference](https://developers.google.com/machine-learning/glossary#agentic) ## NVIDIA NVIDIA develops processors, networking and software used for AI, graphics and scientific computing. - [Explore this context](/entities/nvidia) - [Reference](https://blogs.nvidia.com/) ## Editorial provenance Version 1. Publication ID: 268e09db-5934-4a7e-8f78-40da768374d7. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-05T17:31:23.097Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 874bad201578a9b521bf37a4ab60d60e6b9729615079b84eca2a97f35711798f. Source text hash: b10d79fff8600e8faca4ae0010cc767e3c4fff3e610fddd44c16073f4b6f19a7. ### passed — 2026-11-05T17:31:23.097Z — anthropic/claude-opus-5.5 Publication: 268e09db-5934-4a7e-8f78-40da768374d7 attribution — pass: The reading consistently attributes the architecture, performance figures, enforcement claims and partner activity to NVIDIA's announcement, and labels the performance figures as NVIDIA's stated claims rather than independent evidence. > GTC Berlin—NVIDIA today announced the NVIDIA OpenShell Reference Architecture summary — pass: The title, summary and question faithfully convey the source: NVIDIA bundles hardware and software to contain, monitor and enforce limits on AI agents. Partner scope is described carefully, including exploration versus production. Terms such as DPUs and agent harnesses are kept exact for tooltips; the central idea stays readable. Optional repair: Jensen Huang's framing that agents think on Rubin, act on Vera and are governed on BlueField could restore the announcement's voice, but its omission does not distort meaning. > agents can think on Rubin, act on Vera and be governed on BlueField. date — pass: originalPublished of 2026-10-21 matches the feed entry for this GTC Berlin press release, which says NVIDIA announced it today. There is no evidence of an earlier first publication. support — pass: Each claim matches its excerpt. The 30x throughput-per-megawatt and up-to-20% capacity figures are summarized without overstating them. The limitations correctly note that the source gives no deployment results, no independent evaluation and no definition of the software boundaries. The question follows from that gap without asserting a failure. > NVIDIA Sentry, powered by DOCA and running on BlueField-4, boosts agent capacity by up to 20%. media — pass: The selected media is the verified direct link to the original NVIDIA press release, which is the most direct way to access the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/9ec7aa462c6e54ad62970f0e2f4c18b2a78f5e6cb19baeacb23cc6fa3ec30f41/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/nvidia-announces-openshell-reference-architecture-for-ai-agents-9ec7aa46.json)