Nvidia announces NemoClaw to run AI agents with security controls

In short: Nvidia introduced NemoClaw, an open-source software stack that helps people install and run “AI agents” with added privacy and security controls.

What happened

Nvidia announced NemoClaw on March 16, 2026, during its GTC conference. The company says NemoClaw is built on OpenClaw, an agent platform for running autonomous AI helpers that can keep working in the background.

NemoClaw is designed to make setup simple. Nvidia said it can install its Nemotron AI models and a new runtime called OpenShell with a single command (a one line instruction you paste into a computer to start an install).

A big focus is security. Nvidia says NemoClaw adds privacy controls, policy-based guardrails (rules you set for what the agent is allowed to do), and an isolated sandbox (like a locked room where software can run without freely touching the rest of your computer). Nvidia says this is meant to support “always-on” agents, which it calls “claws,” across cloud services, company servers, and Nvidia RTX PCs and workstations, plus DGX systems.

Nvidia also said NemoClaw uses its NVIDIA Agent Toolkit and includes a “privacy router.” In simple terms, it can help an agent use a local model on your device for sensitive work, while still calling a cloud model when needed, and it applies network and data rules while doing so.

Why it matters

As AI agents get more capable, they also get more access, to files, apps, and the internet. NemoClaw is Nvidia’s attempt to make those agents easier to run while putting clearer limits around what they can see and do, which could matter for both personal use and workplaces.

Source: TechCrunch AI