NVIDIA is handing the governance of OpenShell, its open-source runtime for securing and governing autonomous AI agents, to the Cloud Native Computing Foundation, putting the guardrails for agentic AI under neutral community control.

The contribution was announced on 10 November at KubeCon + CloudNativeCon North America in Salt Lake City. OpenShell joins the CNCF as a Sandbox project, the foundation's earliest stage, and the code is moving from NVIDIA's GitHub repository to a neutral organisation, ai-openshell. The CNCF says the project will be governed by the organisations building it into their products, with a public request-for-comment process, the CNCF code of conduct and explicit vendor-neutrality commitments.

What OpenShell does

OpenShell sits between an AI agent and everything the agent can touch. It applies deterministic controls as agents use tools, access inference services and run workloads, enforcing policies on what files, networks, processes and provider credentials an agent may reach. Agents run in kernel-isolated sandboxes, with declarative YAML policies, and the runtime is adopting Cedar, itself a CNCF project, as its policy language.

NVIDIA says hundreds of organisations have put OpenShell to use and the project has received tens of thousands of contributions in less than a year since launch. Core contributors include NVIDIA, Red Hat, Docker and Canonical. The runtime is licensed under Apache 2.0, so organisations can use, modify and build commercial products on it without paying NVIDIA a licence fee.

Part of a wider Kubernetes push

The OpenShell move came alongside other open-source announcements from NVIDIA at KubeCon. The company is also contributing to Agent Substrate, a Google-originated framework for running long-running agent workloads at scale on Kubernetes that has been accepted into the CNCF Sandbox, and the two projects are being integrated so enterprises can secure and govern agents across clouds and on-premises systems. HPCwire reports that NVIDIA also released three open-source components for its DSX AI Factory platform, including Kubernetes Engine and Kubernetes Recipes in public preview.

Why does this matter now? AI agents are moving from demos to real work: booking travel, changing records, moving money and calling other systems on a user's behalf. Each step widens what can go wrong if an agent is tricked, misreads an instruction or is given too much access. A runtime that enforces policy before the agent acts, rather than logging what it did afterwards, is the layer enterprises have been asking for as they put agents into production.

Sandbox status is only the first step: OpenShell must still earn its way towards incubation, with reproducible builds, conformance tests and security reviews, and the transition to the new organisation is still under way, with code, issues and pull requests remaining at NVIDIA's repository during the move.