NVIDIA introduces infrastructure-level security model for autonomous AI agents
Enterprise adoption grows as NVIDIA integrates security partners for AI agent runtime protection.
OpenShell, an open-source runtime introduced by NVIDIA, is designed to support the secure deployment of autonomous AI agents within enterprise environments.
According to NVIDIA, OpenShell applies security controls at the infrastructure level rather than within the model or application layer. The runtime ensures that each agent operates inside an isolated sandbox, where system-level policies define and enforce permissions, resource access, and operational constraints.
The company states that such an approach separates agent behaviour from policy enforcement, preventing agents from overriding security controls or accessing restricted data.
OpenShell enables organisations to define and monitor a unified policy layer governing how autonomous systems interact with files, tools, and enterprise workflows.
Additionally, OpenShell forms part of the NVIDIA Agent Toolkit and is complemented by NemoClaw, a reference stack designed to support the deployment of continuously operating AI assistants.
NVIDIA indicates that the system can run across cloud, on-premises, and local computing environments, while maintaining consistent policy enforcement.
The company also reports collaboration with industry partners, including Cisco, CrowdStrike, Google Cloud, and Microsoft Security, to align security practices for AI agent deployment. Both OpenShell and NemoClaw are currently in early preview.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!