Red Hat Integrates Nvidia's OpenShell to Secure Enterprise AI Agents

OpenShell places agents in isolated sandbox environments. Red Hat is working with Nvidia to use OpenClaw agents and OpenShell’s architecture together so that agents’ access to potentially vulnerable information is monitored and secured, according to the interview. The effort follows OpenClaw’s rapid adoption and questions about what autonomous agents can access and do.

“OpenClaw came out with a bang,” said Pete Cheslock, principal AI community architect at Red Hat. “This autonomous agent that can just do everything — it was really shocking to see the ecosystem and the community expand around it. As people started to dive into it … you start to ask the questions that everyone should be asking, especially enterprise, which is, what can this access and what can it do? That’s where tools like OpenShell started answering some hard questions.”

Cheslock and Stu Miniman, senior director of market insights for hybrid platforms at Red Hat, spoke with theCUBE Research Chief Analyst Dave Vellante. They discussed how KubeCon has evolved and how Red Hat is addressing AI security concerns with open-source tools. Red Hat will host an OpenShift Commons Gathering on Nov. 9, where open-source practitioners will hear from industry leaders.

Miniman said Kubernetes is only one part of the cloud ecosystem and its heterogeneity is increasing as AI places demands on multicloud environments. “There’s no better way to start your week than going [to OpenShift Commons],” he said. “We’ve got financial services companies, we’ve got companies that are working on AI, working on virtualization, working on some of those key cloud-native pieces, as well as we’ve got two of our Red Hat general managers. We are a clear leading contributor to all of these and have broad customer adoption.”

The OpenShell community is new but growing quickly, Cheslock said, with contributors from a range of companies. It builds on Kubernetes special interest groups, including an agent sandbox, so that agents can be deployed securely using Kubernetes-native APIs, he said. Engineers deploying the agents still face new material to learn, but they do not have to learn the deployment APIs, according to Cheslock.

Miniman said companies are also using AI-powered security tools to test for vulnerabilities at greater speed and scale than human teams can, helping identify and patch gaps before attackers find them. “We’re trying to make sure that customers can feel secure and understand and use these tools,” he said. “That’s front and center. There’s even more effort being put on security than ever, which has always been a focus of Red Hat … it’s that trust, compliance and security. If we make sure that layer is as locked down and reliable and trusted as possible, everything else comes secondary.”