OpenShift 4.22 Adds Controlled AI Agent Access and Distributed Training Support
As artificial intelligence advances rapidly, enterprise demand for deploying and managing AI applications across hybrid clouds is rising sharply. Open-source software company Red Hat continues to modernize OpenShift, its enterprise Kubernetes platform, to address challenges involving resource orchestration, security, privacy and efficient AI model training. These capabilities are critical to helping enterprises operate next-generation AI applications securely.
Red Hat released OpenShift 4.22 in July 2026, introducing MCP server and gateway capabilities as a technology preview for the first time. The features allow AI agents to securely access cluster resources under strict permission controls. The JobSet Operator, designed to coordinate distributed AI training, also became generally available in this release. Together with enhanced confidential-computing and autoscaling capabilities, it significantly improves data security.
All Coverage
1 original reportsThe Backstory
The history behind this eventRed Hat Launches OpenShift 4.21 to Boost AI, GPU Scheduling
Red Hat OpenShift is an enterprise Kubernetes platform designed to run cloud-native applications across on-premise systems and public clouds. The 4.21 release matters because companies increasingly need one operating layer for AI training, containerized services and virtual machines instead of managing separate infrastructure silos. By tightening workload orchestration and accelerator allocation, Red Hat aims to improve utilization of costly GPUs while applying common security, monitoring and governance controls across hybrid-cloud environments.
Red Hat made OpenShift 4.21 generally available on February 3, 2026. Built on Kubernetes 1.34 and CRI-O 1.34, it adds Red Hat build of Kueue 1.2, including Kubeflow Trainer v2 support and visibility into queued workloads. The JobSet Operator reached general availability for distributed computing jobs. Dynamic Resource Allocation also moved to general availability, letting pods request GPUs by attributes such as at least 40 GB of VRAM and specify prioritized device alternatives.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →