Red Hat and Nvidia have expanded their partnership to enhance security for autonomous AI agents and improve workload deployment on upcoming Vera Rubin infrastructure. The collaboration centers on the Red Hat AI Factory with Nvidia, which integrates agent runtime controls and model inference software across hybrid cloud environments.
The security architecture utilizes Nvidia’s Sentry framework running on BlueField data processing units. By leveraging DOCA Argus software, the system monitors host memory and network activity from a separate hardware domain, allowing for the isolation of compromised agents in milliseconds without requiring host-based monitoring software.
The companies are integrating Nvidia’s OpenShell Reference Architecture into Red Hat platforms. This setup separates AI workloads from networking and security services, building on existing BlueField-3 support in Red Hat OpenShift 4.22 to protect infrastructure even if a host is compromised.
New infrastructure updates include support for Nvidia Nemotron models, such as Nemotron 3.5 Super VL, within the vLLM engine. Red Hat Enterprise Linux for Nvidia 26.03 now supports Vera CPUs, ensuring compatibility with the Vera Rubin platform as it becomes available.
The joint reference designs incorporate Model-as-a-Service access controls, an AI Gateway, and inference technologies like Nvidia NIM microservices. These tools are currently available through partners including Cisco Systems, Dell Technologies, and Lenovo Group.
