Red Hat, Nvidia build OpenShell secure agent runtime

The KubeCon + CloudNativeCon NA gathering in Salt Lake City arrives as Kubernetes has become the default container orchestration standard across the enterprise. The open question the cloud-native community is bringing to the event is whether Kubernetes will also anchor the next generation of AI workloads in production, particularly as AI agents move from answering questions to executing code and running business systems.

That shift is what OpenShell is aimed at. Rather than trusting a model to follow rules on its own, the project places enforcement in the environment where the agent operates, and one component runs agents in isolated sandboxes with limits on the files they can reach. Red Hat's Pete Cheslock tied this to core Kubernetes special interest groups, noting the Agent Sandbox lets teams deploy agents securely using Kubernetes APIs. The project's acceptance into the CNCF as a sandbox project in September gives it a home within the ecosystem.

The security emphasis reflects the body's data: 41.3% of organizations cite increased vulnerability risk as their top security challenge, which theCUBE Research's Paul Nashawaty attributes to faster CI/CD cycles. Container use is also broad, with 51% of organizations reporting that 51–75% of their applications now run in containers. Against that backdrop, Red Hat and IBM's $5 billion Project Lightwell commitment to securing open-source software, unveiled in May, sits alongside the OpenShell work as part of the same push.