Red Hat has expanded its co-engineered Red Hat AI Factory with NVIDIA, adding security, governance and lifecycle tools designed to help enterprises run autonomous, long-running AI agents in production rather than confine them to pilots.
The updates were unveiled at Red Hat Summit in Atlanta in May 2026, where the company positioned the platform as a way to close the gap between AI experimentation and day-to-day operations. Built on Red Hat AI Enterprise and NVIDIA AI Enterprise, the factory is pitched as a unified foundation for production-grade agentic AI — software systems that can act on their own over extended periods rather than simply responding to one-off prompts.
Central to the release is OpenShell, an open source project founded by NVIDIA that provides a sandboxed runtime for autonomous agents. It is meant to give organizations a single policy layer governing how agents run, what they can access and where their inference requests are routed, with the stated goal of safer tool use and auditable operations. Red Hat said it is integrating OpenShell with its full AI stack and contributing to the upstream project to help standardize how agents are managed across hybrid cloud environments.
On security, the platform adds hardware-level protection through confidential computing. Now offered as a technology preview, confidential containers run with NVIDIA Confidential Computing inside Red Hat OpenShift sandboxed containers, an approach designed to contain a compromised agent so that a single breach does not spread to others. Red Hat is layering that with a zero-trust architecture drawing on SELinux, FIPS compliance and NVIDIA’s DOCA-based runtime protections, and framed the controls partly as a response to tightening rules such as the EU AI Act.
The factory also folds in Red Hat AI 3.4, including a governed Model-as-a-Service option delivered through the Red Hat AI gateway. That gives developers access to curated models — among them NVIDIA’s Nemotron — through standard, OpenAI-compatible interfaces. For oversight, the platform uses MLflow to trace model calls, tool execution and reasoning steps, so teams can audit how an agent arrived at a given result.
Underpinning the software is Red Hat Enterprise Linux for NVIDIA 26.01, now generally available, which the company said delivers Day 0 support for new NVIDIA chips. The first release supports NVIDIA’s Blackwell architecture, with work already underway on the upcoming Vera Rubin platform, and NVIDIA’s Run:ai workload tool is now included for customers. Red Hat and NVIDIA are also shipping validated blueprints and quickstarts for common patterns such as enterprise research and retrieval-augmented generation.
Justin Boitano, NVIDIA’s vice president of enterprise AI platforms, said every company “will need an AI factory to build, deploy and govern digital workers at scale.” Chris Wright, Red Hat’s chief technology officer, framed the collaboration as a way for enterprises to scale agentic AI while retaining architectural control and technical independence across the hybrid cloud.
The push reflects a broader industry race to make AI agents enterprise-ready, a shift in which security, governance and auditability increasingly matter as much as raw model performance. Red Hat said the AI Factory with NVIDIA software is available now, with the latest updates and Red Hat AI 3.4 rolling out in the weeks following the announcement.
