Nvidia CEO says OpenClaw agent is ‘next ChatGPT’

Jensen Huang, Nvidia CEO, said in a “Mad Money” interview on CNBC that OpenClaw was “definitely the next ChatGPT” and described it as an open-source autonomous agent platform that could complete tasks with minimal user input.

OpenClaw enables agents to make decisions and act rather than only answer questions.

Nvidia announced NemoClaw on March 15, an enterprise-grade version that layers its software stack and tools on OpenClaw to add security and scalability.

Huang said developers could create an agent in one line of code and gave a kitchen design example where an agent studied images and iterated on designs autonomously.

The rapid rise of autonomous agents like OpenClaw has raised concerns around privacy, security, and control as they grow more capable.

Nvidia said NemoClaw would add guardrails such as privacy protections, oversight tools, and enterprise-grade security to support safer deployment at scale.

🔗 Source: CNBC

🧠 Food for thought

Implications, context, and why it matters.

Security gaps that made NemoClaw feel timely for businesses

OpenClaw took off with developers, topping 250,000 GitHub stars in about four months, fueled partly by interest in a free personal AI assistant that connects with WhatsApp and Slack 1.
Fast uptake also brought security worries since OpenClaw can collect high-privilege credentials for email, cloud services, and messaging apps, plus those secrets may sit in plain text configuration files 2.
A security audit of one marketplace found almost 12% of agent “skills” (third-party add-ons that extend what an agent can do) were malicious, built to grab credentials and browser data using trojans tied to the AMOS stealer family (a known strain of credential-stealing malware) 3.
With this setup, one hostile skill can reach every system the agent touches, which turns a personal assistant into an entry point for attackers 2.

Nvidia frames NemoClaw as a security layer for rolling out agents at work

Nvidia presents NemoClaw as an infrastructure layer, with CEO Jensen Huang saying each company needs an “agentic systems strategy” (a company-wide plan for deploying autonomous AI agents) like earlier shifts to Linux, the web, and Kubernetes 4.
OpenShell, which Nvidia calls the isolation component for agent execution, runs agents inside an “out-of-process” security sandbox (a separate, restricted runtime environment) that sets rules for the agent’s surroundings, which blocks a compromised agent from changing its own safety controls 5.
Nvidia says NemoClaw can run “anywhere” and sells it as an open-source stack that adds privacy and security controls to OpenClaw; WIRED also reported access regardless of whether customers run Nvidia chips 6.
The company also steers always-on agent use toward GeForce RTX (consumer and workstation graphics cards) plus DGX systems (Nvidia’s enterprise AI server platforms), even while calling NemoClaw broadly deployable 7.