SUSE adds NVIDIA security controls for private enterprise AI agents

Enterprises seeking to run AI agents inside their own infrastructure are getting new controls over what those agents can access and send outside. SUSE’s October 21, 2026 update to SUSE AI Factory with NVIDIA integrates NVIDIA’s OpenShell Reference Architecture and introduces a dedicated Linux offering, scheduled for general availability in November.

The coordinated release addresses two distinct jobs: keeping autonomous agents within company-defined rules and maintaining the infrastructure beneath them. SUSE says the package gives developers a foundation for building applications, platform teams a way to deploy them, and enterprise IT control over their operation across private and hybrid environments.

Checking requests as agents work

SUSE describes the security approach as zero trust: agent traffic is continually inspected, and access depends on defined policies. Supervisor proxies check requests and block unauthorized network activity to help prevent code or data from leaving without permission. The platform also uses Kubernetes sandboxing—isolated execution environments—and API gateways alongside OpenShell to govern agent execution.

The underlying reference architecture divides model processing, agent execution and security monitoring across different NVIDIA components. SUSE describes the division this way:

• Model processing: NVIDIA Vera Rubin handles inference, the work of running a model to produce outputs.
• Agent execution: NVIDIA Vera CPUs run OpenShell, providing the environment for agents to carry out their work.
• Security monitoring: BlueField-4 data processing units run NVIDIA Sentry, powered by NVIDIA DOCA, for host-independent monitoring and policy enforcement in silicon.

For data control, SUSE says agent code can execute within the corporate perimeter, with policies restricting transfers of proprietary datasets and intellectual property outside private infrastructure. These are the company’s stated safeguards, rather than a guarantee that every deployment will prevent unauthorized transfers.

By combining enterprise Linux engineered for bleeding-edge hardware with active agent security, SUSE AI Factory with NVIDIA OpenShell equips IT teams to scale AI rapidly while keeping proprietary data, system availability, and budgets strictly under their control.

Dr. Thomas Di Giacomo, SUSE chief technology and product officer

A Linux base tied to NVIDIA’s hardware cycle

SUSE Linux Enterprise Server for NVIDIA, or SLES for NVIDIA, uses a SLES 16 base and is designed to support NVIDIA’s latest hardware, including Vera Rubin infrastructure. SUSE is positioning it for rack-scale systems, with the operating system kept aligned to NVIDIA’s hardware release cycle rather than treated as a separate integration task.

The support promise spans more than Linux itself. SUSE says it will provide a single technical-support source backed by continuous automated validation across the operating system, Linux kernel, GPU drivers and Kubernetes runtime. It also says developers should be able to move into production without rewriting code or changing platforms. Those compatibility benefits remain vendor claims.

A government trial, with infrastructure-based costs

One named test is underway at Mexico’s Guanajuato Secretariat of Finance. According to SUSE, the agency is evaluating AI-driven public services and internal workflows through a proof of concept. It is also testing deployment on high-performance GPU infrastructure. The work is an evaluation, not an announced production rollout of those public services.

The cost argument also rests on running models privately. SUSE says NVIDIA NIM microservices allow private model serving without third-party public API charges for each token processed. Its proposed alternative is capacity-based infrastructure investment: spending on the computing resources that serve the models. That describes a different charging model, not evidence of lower total costs.