SUSE adds NVIDIA agent safety to its closed AI platform
SUSE tells that the new part keeps AI agent code in the network of the company and stops the transfer of data out. It gave no test results.
On 23 October 2026, SUSE, a Linux and open infrastructure company, told of a new part for its SUSE AI Factory platform with NVIDIA. The part is the NVIDIA OpenShell Reference Architecture, a setup to operate AI agents in a closed area. SUSE names 6 approved software partners but gave no speed numbers or test results.
How it works
Supervisor proxies are software between an agent and the network. They check agent traffic against policies. They stop requests that break the policies.
The setup uses NVIDIA chips named Vera Rubin for model work. It uses Vera CPUs for agents and BlueField-4 cards for monitoring that does not use the host server.
The proof
SUSE gave no test results. The finance office of Guanajuato, a part of Mexico, tests the platform in a trial. SUSE tells that its closed model serving removes the cost of outside AI tools that charge for each use. Gartner tells that by 2027, 35% of countries will be locked into AI platforms for one region.
What is not known
SUSE gave no price and no date for the hardware. All information on safety and costs is from SUSE.
Insight: NVIDIA is making agent safety a chip feature
In one week, NVIDIA had confidential servers on 21 October, European sovereign standards on 19 October, and at this time agent monitoring in its cards. AWS made its agent sandbox on 16 October. It reads as a contest to control the agent wall, and monitoring put into NVIDIA chips can attach the safety layer to NVIDIA hardware. Look for other Linux vendors to connect.
Sources
Posted
