Red Hat expands NVIDIA tie-up on secure agentic AI

Red Hat has expanded its collaboration with NVIDIA on agentic AI across hybrid cloud environments, focusing on security, model inference and hardware support.

The latest changes will be delivered through Red Hat AI Factory with NVIDIA, a joint offering for organisations building and running autonomous AI agents.

One part of the update focuses on runtime security for AI agents. Red Hat is working with NVIDIA on the OpenShell Reference Architecture, designed to set boundaries and guardrails for agents outside the host operating system.

The approach uses OpenShell, an open source agent sandbox runtime project to which both companies contribute. The project is intended to provide standard guardrails, operational visibility and runtime safety for agents in complex enterprise environments.

Security focus

The announcement reflects a broader shift in corporate AI deployments as companies move from chatbots and assistants to systems that can act with greater autonomy. Those systems create new demands for controls over where agents run, what they can access and how their behaviour is monitored.

NVIDIA also described the issue as one of separating controls from the agent itself. The aim is to place those boundaries in the underlying runtime and infrastructure rather than in the software agent alone.

"An agent shouldn't be enforcing its own boundaries - those controls belong outside the agent, in the runtime and the infrastructure beneath it. NVIDIA's expanded work with Red Hat combines a secure agent runtime, governed inference and deep hardware support to give enterprises a trusted, high-performance blueprint for agentic AI across the open hybrid cloud," said John Fanelli, Vice President, Enterprise Software, NVIDIA.

Inference design

Another element of the update is a reference design for model inference and governance. It combines Model-as-a-Service access control and AI Gateway functions with inference tools including vLLM and llm-d, as well as NVIDIA Dynamo, NVIDIA NIM microservices and GPU orchestration.

According to the companies, the reference implementation is based on the NVIDIA Inference Hub design. It is intended to provide self-service model discovery, access policies, rate limiting and subscription-based entitlement tracking.

The setup separates governance from execution so organisations can manage model access and quotas while running inference workloads across hybrid cloud environments. It is also intended to support cost attribution and operational control.

The collaboration also includes support for NVIDIA Nemotron models in vLLM and Red Hat AI Factory with NVIDIA. Red Hat highlighted full support for NVIDIA Nemotron 3.5 Super VL and said its LLM compressor can be used with quantisation methods intended to reduce inference latency while preserving accuracy.

Hardware support

A third strand of the work covers hardware enablement. Red Hat plans Day 0 support across its open hybrid cloud portfolio for upcoming NVIDIA architectures, including NVIDIA Vera Rubin and NVIDIA Vera CPUs.

Red Hat Enterprise Linux for NVIDIA now also supports NVIDIA Vera CPUs. Red Hat described the operating system edition as a version of Red Hat Enterprise Linux tailored with NVIDIA.

This hardware alignment is aimed at organisations seeking operating system support and platform compatibility when new NVIDIA infrastructure becomes available. The companies linked that need to more demanding AI agent workloads, which can require continuous reasoning loops, tool orchestration and sandbox execution.

Use cases and reference designs tied to the collaboration will be available through Red Hat's partner ecosystem, including Cisco, Dell Technologies and Lenovo.

Joe Fernandes, Vice President and General Manager, AI Business Unit, Red Hat, said the work is intended to address the operational demands of deploying autonomous AI systems in business settings. "Enhancing the security of agentic AI will fuel the next major evolution in enterprise computing, but scaling autonomous agents requires trust, safety and seamless infrastructure integration. By deepening our collaboration with NVIDIA across open source communities like OpenShell, delivering optimised inference architectures and driving day-zero hardware enablement, we are giving IT decision-makers the trusted foundation they need to confidently deploy AI agents at scale," said Fernandes.