NVIDIA donates OpenShell agent security runtime to CNCF

NVIDIA announced on November 10, 2026, at KubeCon + CloudNativeCon, that it has donated OpenShell, its open-source runtime for governing what autonomous AI agents can do when using tools, calling inference services and executing workloads, to the Cloud Native Computing Foundation (CNCF) as a Sandbox project. The announcement was published on NVIDIA's blog, alongside three new infrastructure components for running AI agent and inference workloads on Kubernetes.

What's changing

• OpenShell, NVIDIA's open-source runtime for governing agent tool use, inference calls and workload execution, becomes a CNCF Sandbox project.
• NVIDIA says hundreds of organizations already use OpenShell, with tens of thousands of contributions in under a year since launch.
• NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR) entered public preview as open-source tooling for managed Kubernetes on GPU infrastructure, part of the DSX AI Factory Platform, with integration partners including Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud, vCluster, Firmus and SK Telecom.
• NVIDIA Snapshot restores checkpointed inference replicas in seconds without a full model reload; design partner Crusoe cut GLM-5.3 workload readiness time from about 8 minutes to 31 seconds, and Saturn Cloud reports vLLM startup times improved from 2 to 8.5 minutes down to 6 to 34 seconds.
• NVIDIA is also contributing to a Google-led framework called Agent Substrate for running long-running agent workloads on Kubernetes, integrating it with NVIDIA Dynamo and the KAI Scheduler.

What it means for AI builders

For developers and platform teams, the shift suggests that governance and fast-restart infrastructure for agentic AI workloads on Kubernetes is turning into a shared, open-source layer across major cloud-native vendors. Putting OpenShell under CNCF governance gives agent-security tooling a neutral home that competitors can build on together, while NVKE, NVKR and Snapshot address the operational side, restarting inference workloads in seconds rather than minutes, so that teams deploying and securing AI agents at scale have a common set of building blocks instead of proprietary, vendor-specific ones.

Sources

• NVIDIA Donates OpenShell to CNCF, Advances Open Infrastructure for AI Agents (NVIDIA Blog, November 10, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.