Nvidia wants in on the OpenClaw craze

OpenClaw is a platform for building autonomous agents that can independently complete tasks, delegate work to purpose-built subagents, and reach into a user's local file system. NemoClaw addresses a structural vulnerability in that design: The agents need sweeping permissions over files and accounts to do their jobs, which means a misbehaving or compromised agent can do real harm. CrowdStrike $CRWD's chief technology officer described agents that delete emails or surrender passwords as known failure modes, according to The Wall Street Journal.

NemoClaw, announced at Nvidia's annual GTC conference, adds a contained sandbox environment and policy-based guardrails for network access, data privacy, and security. It can run open models locally — including Nemotron — or route queries to frontier cloud models through a privacy router. Nvidia said NemoClaw works with any coding agent and is compatible with dedicated GeForce RTX PCs and laptops, RTX PRO workstations, DGX Station, and DGX Spark AI supercomputers.

"Now it's enterprise ready," Nvidia CEO Jensen Huang said. At GTC, Huang called OpenClaw "the operating system for personal AI," putting it in the same category as Mac and Windows in terms of platform significance, according to CNN.

OpenClaw creator Peter Steinberger, who was recently hired by OpenAI, said in a statement that the partnership with Nvidia is aimed at building "the claws and guardrails that let anyone create powerful, secure AI assistants."

The NemoClaw announcement was part of a broader set of agent-focused announcements at GTC, where Huang framed autonomous AI agents as the next major phase of enterprise computing.