Red Hat, the world’s leading provider of open source solutions, announced new milestones in its co-engineered Red Hat AI Factory with NVIDIA software platform—advancing enterprise adoption of AI and autonomous, long-running agents. The latest updates introduce enhanced security capabilities for continuously operating agents and deliver ready-to-deploy solutions that help organisations move from experimentation to production faster and with confidence. Powered by the consistency and integrated AI platform capabilities of Red Hat AI Enterprise along with the open models, advanced tools, and frameworks within NVIDIA AI Enterprise, this solution provides a unified, scalable foundation for production-grade agentic AI.
Providing a safe runtime for a new class of agents
OpenShell is an open source project founded by NVIDIA that provides a sandboxed runtime for autonomous AI agents, built to give enterprises stronger policy controls, safer tool use, and auditable operations. It governs how agents execute, what they can access, and where inference is routed. With OpenShell, organisations gain a single, unified policy layer to control and monitor how autonomous agents operate.
Joint engineering work is underway to integrate OpenShell with Red Hat’s full-stack AI platform to help maintain oversight and policy at the infrastructure level. Also, as a key contributor to the OpenShell upstream open source project, Red Hat is helping to standardise how autonomous agents are managed across the hybrid cloud.
AI sovereignty, security and architectural control with confidential containers
Beyond the software-defined governance, the Red Hat AI Factory with NVIDIA extends the trust boundary and enforces policies at the hardware level with confidential computing. This is designed to protect against the runtime compromise of agents, even if another agent is breached. Now available in technology preview, users can run confidential containers with NVIDIA Confidential Computing as part of Red Hat OpenShift sandboxed containers.
This hardware-enforced protection is further bolstered by a zero-trust architecture that integrates features like SELinux, FIPS compliance, and NVIDIA DOCA-based runtime protection. By layering these controls from the core data center to the edge, the Red Hat AI Factory with NVIDIA helps enterprises protect critical assets in highly distributed environments and address evolving security and governance directives, including those introduced by frameworks such as the EU AI Act.
Unified lifecycle management and Model-as-a-Service
Red Hat AI Factory with NVIDIA software now incorporates the breadth of innovations in Red Hat AI 3.4. This includes a new, governed Model-as-a-Service (MaaS) experience delivered via the Red Hat AI gateway, providing developers with streamlined access to curated models—such as NVIDIA Nemotron—through standard OpenAI-compatible interfaces. To support operational oversight, the factory integrates production-ready lifecycle management powered by MLflow, enabling end-to-end tracing of LLM calls, tool execution, and reasoning step. This unified framework bridges the gap between developer velocity and operational stability, allowing organisations to audit how an agent arrived at an outcome.
