NVIDIA and NemoClaw: Europe Faces a Major Regulatory Challenge

At the GTC 2026 in San José, NVIDIA made a major announcement with the launch of NemoClaw, a platform that could very well become the standard for autonomous agents in business. NVIDIA's CEO, Jensen Huang, presented this ambitious vision, emphasizing that the company is no longer limited to manufacturing chips for AI but aspires to become the software foundation for these agents.

An Impressive Demonstration

In his speech, Jensen Huang welcomed Peter Steinberger, the creator of OpenClaw, for a striking demonstration. Huang compared OpenClaw to iconic technologies such as Linux, Kubernetes, and the HTTP protocol, describing it as a true operating system for AI agents. The highlighted features include resource management, file access, calls to language models, task scheduling, and the creation of sub-agents. Huang stressed the importance for every company to define its own “OpenClaw strategy.”

Security Challenges

Huang also addressed security concerns related to autonomous agents, particularly their ability to access sensitive information, execute code, and communicate externally. He presented NemoClaw as a solution to these challenges, with a simplified installation via a single command. The platform deploys the OpenShell runtime, an environment that frames the behavior of agents through configurable safeguards for permissions, network access, and privacy policies. Agents use the open models Nemotron locally, while a privacy router filters calls to cloud models. NVIDIA describes this as a “reference design” that any SaaS company can connect to its own security policy engine.

An Economic Transformation

Huang positioned NemoClaw not just as a product but as a shift in the business model for the software industry. According to him, every SaaS publisher will become a “GaaS”, a provider of AI agents as a service. Companies will no longer sell tools but specialized agents that clients can “rent.” To illustrate this idea, Huang proposed that every NVIDIA engineer receive an “annual token budget” in addition to their salary, allowing access to AI agents that enhance their productivity. However, this concept of a “token budget” has no equivalent in current salary grids in France, and no collective agreement provides for it.

Regulatory Challenges in Europe

The real issue lies in data management by these agents. An active agent can continuously access files and emails, acting without human validation. Past incidents, such as one involving a Meta employee, highlight potential risks. NemoClaw promises controls, with OpenShell isolating agents in a sandbox and the privacy router filtering sensitive data. However, NVIDIA acknowledges that the platform is still in alpha and warns developers to expect imperfections.

In Europe, the GDPR imposes strict obligations on the automated processing of personal data, and the AI Act will add constraints for high-risk autonomous systems. However, NVIDIA has not detailed specific mechanisms for the European framework, and the term “GDPR” does not appear in either the keynote or the technical documentation. An AI agent operating continuously on a workstation accesses third-party emails. If excerpts are transmitted to a cloud model, this constitutes personal data processing, for which the company deploying the agent is responsible.

The gap between the vision presented in San José and European regulatory maturity is immense. While Huang hopes that every company will have an OpenClaw strategy, Europe must first ensure that every company has a control strategy.