Red Hat says it has expanded its collaboration with NVIDIA to support security-focused “agentic” AI deployments across hybrid cloud environments, using Red Hat AI Factory with NVIDIA.
The companies said the updates are aimed at enterprises moving from generative AI pilots to more autonomous workflows, where software agents act on behalf of users and require additional controls to reduce the risk of unsafe or unintended actions.
One element of the expanded work is support for NVIDIA’s OpenShell Reference Architecture, which is intended to place enforcement controls outside an AI agent and establish runtime boundaries and guardrails independent of the host operating system. Red Hat said it is integrating OpenShell-related capabilities through Red Hat AI Factory with NVIDIA, and contributing to OpenShell as an open source project.
The companies also described a joint reference design focused on scaling inference for agent workloads, incorporating Model-as-a-Service access control, an AI Gateway, and inference components including vLLM and llm-d, alongside NVIDIA tooling such as Dynamo and NIM microservices. The reference design is intended to provide features such as self-service model discovery, fine-grained access policies, rate limiting and subscription-based entitlement tracking, with governance separated from execution to support hybrid cloud deployments.
Red Hat said it is also providing support for optimised NVIDIA Nemotron 3.5 Super VL models in vLLM and Red Hat AI Factory with NVIDIA. It cited model quantisation via Red Hat’s LLM compressor and the use of FP8 and DSpark speculative decoding checkpoints as techniques to reduce inference latency while maintaining model accuracy.
On the infrastructure side, the announcement includes “Day 0” enablement for upcoming NVIDIA platforms, including NVIDIA Vera Rubin. Red Hat said its specialised Red Hat Enterprise Linux for NVIDIA edition will also support NVIDIA Vera CPUs, which NVIDIA positions for agentic reasoning workloads.
All joint use cases and reference designs are available through Red Hat’s partner ecosystem, including Cisco, Dell Technologies and Lenovo, the companies said.
“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,” said Joe Fernandes, vice president and general manager, AI Business Unit, Red Hat.
John Fanelli, vice president, Enterprise Software, NVIDIA, said: “An agent shouldn’t be enforcing its own boundaries — those controls belong outside the agent, in the runtime and the infrastructure beneath it.”
