NVIDIA NemoClaw
Policy-based guardrails for autonomous AI agents — the missing security layer
NemoClaw is NVIDIA's official security and privacy stack for OpenClaw agents, announced at GTC 2026. It installs in a single command and adds NVIDIA OpenShell — an open-source runtime that enforces policy-based guardrails, sandboxes AI models, and adds data privacy protections. Built alongside CrowdStrike, Cisco, and Microsoft Security. Runs on any platform: cloud, on-premises, RTX PCs, DGX Station, DGX Spark.
Use Cases
- Enterprise deployment of autonomous OpenClaw agents
- Policy-based access control for agent permissions
- Compliance and audit trails for AI decision-making
- Isolation of sensitive data from agent access
- Runtime enforcement of operational guardrails for ZHCs
Key Features
Policy-Based Guardrails
Organizations define what agents can and cannot do. OpenShell enforces those policies at runtime — not just at the prompt level.
Model Sandboxing
Keeps AI models isolated from sensitive systems unless explicitly authorized. Prevents unauthorized data access or exfiltration.
Data Privacy Protections
Built-in controls for what data agents can read, write, or transmit. Essential for ZHCs handling customer or financial data.
Enterprise Security Integration
Compatible with CrowdStrike, Cisco, and Microsoft Security. Integrates with existing enterprise security stacks rather than replacing them.
Single-Command Install
One command installs both Nemotron models and OpenShell runtime. Works across cloud, RTX PCs, DGX Station, and DGX Spark.