Nvidia donates AI agent security runtime OpenShell to CNCF

Summary

• Nvidia is contributing OpenShell, a runtime that governs what autonomous AI agents can do, to CNCF.
• OpenShell enters the CNCF Sandbox, governed by the organizations embedding it rather than by Nvidia alone.
• Enterprises running agents gain a neutral, inspectable control layer instead of a vendor-owned one.

The latest

Nvidia is handing its agent security runtime, OpenShell, to the Cloud Native Computing Foundation as a Sandbox project, placing control over what autonomous agents may do inside enterprise systems under neutral governance. The company announced the donation at KubeCon + CloudNativeCon alongside contributions to Google's Agent Substrate project and three new Kubernetes components for its DSX AI Factory Platform.

Details

• What OpenShell does: Nvidia said the runtime applies deterministic controls to agent activity, letting teams enforce policy as agents call tools, reach inference services and execute workloads. Its extensible architecture allows organizations to plug in their own compute, inference providers, credential storage and monitoring systems.
• The governance change: As a CNCF Sandbox project, OpenShell will be governed by the organizations building it into their products rather than by Nvidia alone, according to the company. CNCF is a nonprofit under the Linux Foundation and the host of KubeCon.
• Adoption to date: Nvidia said hundreds of organizations have deployed OpenShell and the project has drawn tens of thousands of contributions in under a year since launch. The company did not break down which organizations are running it in production.
• The Linux Foundation view: Jim Zemlin, CEO of the Linux Foundation, said giving agents secure environments to operate in is a new infrastructure challenge for the cloud-native community, and described the donation as "a shared foundation to build on" for developers shaping agent security.
• The Google project: Nvidia is also contributing to Agent Substrate, a Google-led open source framework for running long-lived agent workloads at scale on Kubernetes that has been accepted into the CNCF Sandbox. It preserves agent environments between tasks while releasing compute when agents sit idle.
• The integration work: Nvidia said it is making Agent Substrate work with Nvidia Dynamo for model inference and KAI Scheduler for shared cluster resource allocation. Nvidia and Google are separately building an OpenShell–Agent Substrate integration so enterprises can govern agents across clouds and on-premises systems.
• New Kubernetes tooling: Nvidia Kubernetes Engine and Nvidia Kubernetes Recipes, both in public preview, are open source building blocks for delivering managed Kubernetes services on GPU infrastructure. They automate cluster lifecycle management and ship signed, version-pinned images. Nvidia did not give a general availability date for either tool.
• The partner list: Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud and vCluster, along with Firmus and SK Telecom, have aligned with Nvidia on integration plans for the new Kubernetes components.
• The speed numbers: Nvidia Snapshot restores checkpointed replicas in seconds without model reload or warmup. Crusoe tested GLM-5.3 on Crusoe Cloud and cut time-to-ready from roughly eight minutes to 31 seconds; Saturn Cloud reduced vLLM startup from between two and 8.5 minutes to between six and 34 seconds.

Background

Kubernetes, the open source system that schedules applications across large compute clusters, has become the default substrate for AI workloads. Nvidia framed the KubeCon announcements as a response to rising demands on that layer as autonomous agents move into production.

Between the lines

Handing OpenShell to a neutral foundation while wiring it to Nvidia Dynamo and KAI Scheduler lets the company open the governance layer without loosening its grip on the inference and scheduling pieces underneath. The Agent Substrate work points the same way: Nvidia is not originating the orchestration standard, it is ensuring the standard others build runs well on its hardware.

What's next

Watch whether OpenShell graduates beyond CNCF Sandbox status, when Nvidia Kubernetes Engine and Recipes exit public preview, and whether the joint Nvidia–Google OpenShell integration with Agent Substrate ships as announced.

Source: Nvidia Blog