NVIDIA donated OpenShell to the CNCF: what it secures, and what it leaves open when agents deploy to your own cloud
Blog post from Qovery
NVIDIA contributed OpenShell, an Apache 2.0 runtime for isolating AI agents, to the CNCF Sandbox on November 10, 2026, with plans to move it to a neutral project organization and adopt Cedar for policy. OpenShell uses kernel-level controls and declarative rules to restrict agents’ files, processes, system calls, network access and credential exposure, but its Sandbox status and 0.1.x releases mean it should be treated as an evolving tool rather than a production-ready guarantee. The central distinction is that execution sandboxing limits what an agent can access, while deployment governance determines what it can change: if an agent can reach a cloud or Kubernetes API, its effective permissions depend on the credential it uses, and the sandbox may not prevent mistaken or unauthorized production changes. The article therefore recommends pairing sandboxing with per-agent, narrowly scoped credentials, environment-aware deployment policies, human approval for production, disposable preview environments and audit records that identify the agent. It presents Qovery as a complementary deployment-control layer, not a sandbox, offering policy-checked API access and action auditing while applications remain in the customer’s cloud or cluster. Teams are advised to pilot OpenShell, monitor its policy and API changes, and coordinate sandbox controls with deployment permissions.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| Kubernetes | 28 | 3,388 | 294 | 94 | -34% |
| AI Agents | 18 | 4,689 | 923 | 218 | -33% |
| MCP | 15 | 7,404 | 711 | 183 | -38% |
| Platform Engineering | 5 | 1,080 | 220 | 61 | -46% |
| AI Coding Assistant | 2 | 1,395 | 336 | 123 | -29% |
| Agent sandbox | 2 | 61 | 25 | 9 | +154% |
| Developer Experience | 1 | 195 | 133 | 41 | -60% |
| LLM | 1 | 2,788 | 618 | 179 | -58% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.