Diverging from OpenAI's Path! Nvidia Quietly Builds AI Safety Team, Betting on Open-Weight Models

Recent reports from foreign media indicate that Nvidia is secretly forming a new AI safety and network engineering team to enhance defensive capabilities in the era of open-weight models.

According to Business Insider on Thursday, job postings revealed at the end of last month first exposed this team. The positions include Distinguished Engineer (founding technical lead), Security Research Engineer, Evaluation Engineer, and Senior Manager. Their responsibilities involve conducting safety evaluations on AI agents before deployment and developing tools that use AI to automatically fix software vulnerabilities.

In the job descriptions, Nvidia emphasized a 'firm belief': open weights, technical transparency, and broad scientific review are the foundation for the United States to maintain its leadership in AI and strengthen cyber defense.

This move highlights how Nvidia, while betting heavily on open-weight models and autonomous agents, is elevating AI safety to a core strategic priority. Open-weight models publicly release the training parameters that determine how the model operates, while still allowing training data and source code to remain closed.

In recent months, Nvidia has repeatedly advocated for open models, creating a stark contrast with the closed-source approaches of OpenAI and Anthropic.

Last month, Nvidia CEO Jensen Huang made his first public endorsement on the X platform, retweeting an open letter urging U.S. policymakers to support open models. He stated clearly that openness enhances AI safety and strengthens cybersecurity defenses.

As AI evolves from conversational chatbots into agents capable of reading sensitive corporate data, invoking tools, and executing real-world tasks, the level of trust enterprises place in these systems will determine the speed of adoption.

Analysts point out that Nvidia's formation of a dedicated security team serves not only to protect its own GPU ecosystem but also to seize early control over the industry standard of 'security as moat' on the eve of the Agentic AI explosion.