Nvidia's NemoClaw aims to tame the wild west of AI agents with enterprise-grade security

Nvidia is betting big on the future of autonomous AI agents, but with a crucial twist: enterprise-grade security. At its GTC conference in San Jose, the chipmaker unveiled NemoClaw, a comprehensive software stack designed to make the OpenClaw platform safe for corporate use.

The claw revolution and its security nightmare

The terminology might sound whimsical, but the implications are serious. "Claws" - Nvidia's shorthand for software agents - represent AI models given direct access to tools and services, capable of planning, acting, and executing tasks autonomously. Kari Briski, Nvidia's VP of generative AI software for enterprise, explained during a pre-conference briefing that the paradigm has shifted from prompting with "what, how, or why" to commanding with "build, create, or make."

This evolution traces back to January's OpenClaw debut, an open platform that briefly captured social media's imagination by demonstrating how easily automation could be enabled. The project, originally known as Clawd and then Moltbot, showed the world both the potential and the peril of unrestricted AI agents. As Briski noted, "Claws are exciting but they're risky too, because they could access sensitive data, misuse connected tools, or escalate privileges autonomously."

Enter NemoClaw: Security as the new foundation

Nvidia's solution wraps OpenClaw in multiple layers of protection. The NemoClaw stack allows users to install Nvidia's Nemotron models and OpenShell runtime through a single command via the Nvidia Agent Toolkit. This toolkit bundles together models, runtimes, and blueprints specifically engineered for safer, long-running agents.

At the heart of this security approach is OpenShell, which Briski describes as "an open-source safety and security runtime for agents." Think of it as a digital sandbox that limits what AI agents can access and do. "OpenShell provides the missing infrastructure layer beneath claws to give them the access they need to be productive while enforcing policy-based security, network, and privacy guardrails," she explained.

Enterprise-ready AI agents

The timing is strategic. As companies grapple with whether to embrace AI automation, many remain paralyzed by security concerns. Nvidia CEO Jensen Huang has positioned OpenClaw as "the operating system for personal AI," but corporations need more than personal AI - they need controlled, auditable, and compliant systems.

Nvidia's approach addresses these concerns head-on. NemoClaw for OpenClaw supports local computing across multiple platforms: PCs with Nvidia GeForce RTX graphics cards, workstations with RTX Pro, and even high-end systems like the DGX Station and DGX Spark supercomputers. This flexibility means companies can deploy AI agents in environments they control, rather than sending sensitive data to the cloud.

The bigger picture: AI's compute demands

Beyond security, there's a business case driving Nvidia's push. Briski was blunt about the opportunity: "Claws are the new application layer for AI, and they're driving orders of magnitude more demand for compute." As AI agents become more capable and autonomous, they'll require significantly more processing power - exactly what Nvidia sells.

This mirrors the broader industry trend where companies are racing to capitalize on AI's potential while managing its risks. Just as Docker's Sandbox has provided a home for NanoClaw, Nvidia is creating enterprise-grade infrastructure for the next generation of AI applications.

The claw may choose who stays and who goes in Pixar's Toy Story, but in Nvidia's vision, enterprise IT departments get to choose which claws get to play with corporate data - and under what strict rules.