Red Hat and NVIDIA open-source OpenShell, bringing default-deny security controls to AI agents

On November 11, 2026, SiliconANGLE reported that Red Hat and Nvidia had expanded their partnership and open-sourced OpenShell, a project aimed at giving AI agents that move from developer laptops into data centers deny-by-default network and access restrictions. The stated goals are to address prompt injection and data exfiltration risks and to create a shared, enterprise-grade standard for agent security guardrails that do not strip away what makes agents useful.

The following day, November 12, 2026, DEV Community published a technical breakdown of OpenShell, describing it as a policy layer that runs on top of a container or VM runtime rather than replacing the runtime. It provides egress rules based on the binary being invoked, restrictions on methods and paths, and Landlock filesystem isolation inside workloads. The piece illustrates the problem with a coding agent running in a container as a non-root user: the image, compose file, firewall script, and secrets manager each hold only part of the answer, so none can individually list which files, processes, network destinations, and credentials the agent can reach. OpenShell is positioned as the layer that ties those pieces together. So far, public reporting has remained at the level of project introduction and launch, with no follow-up developments disclosed.