What changed
During KubeCon + CloudNativeCon NA, Red Hat and Nvidia announced the development of OpenShell, an open-source secure agent runtime designed for Kubernetes environments. OpenShell aims to provide a secure framework for AI agents by setting boundaries on what they can access and execute, rather than relying solely on the agents themselves to follow rules.
The project, still in early stages, was accepted into the Cloud Native Computing Foundation (CNCF) as a sandbox project in September. It includes features that allow agents to run in isolated sandboxes with file access limitations, a concept actively supported by the Kubernetes community.
What this means in practice
OpenShell reflects growing enterprise interest in using containerized environments for AI applications. According to theCUBE Research, 51% of organizations now run 51–75% of their applications in containers, a sharp increase from just a few years ago. This trend underscores the accelerating shift toward cloud-native technologies.
The project builds on core Kubernetes special interest groups, including the Agent Sandbox, which enables secure deployment of agents using Kubernetes APIs. This integration highlights the increasing focus on Kubernetes-native security tools to manage the risks associated with AI agents.
Limitations and availability
The supplied announcement does not specify the exact release date or availability of OpenShell. It is currently an early-stage project under development for Kubernetes environments. The CNCF sandbox status indicates it is in the early phases of development and may not yet be widely available.
The project is part of a broader effort to enhance security in cloud-native environments, including initiatives like Project Lightwell, a $5 billion commitment by Red Hat and IBM to secure open-source software. These efforts reflect the growing emphasis on security as enterprises adopt AI and cloud-native technologies.
