Treasury Secretary Advocates for Open-Source AI Development

On Tuesday, Treasury Secretary Scott Bessent addressed lawmakers, emphasizing the necessity for the U.S. to create more open-source artificial intelligence (AI) models to remain competitive with China.

During his testimony before the House Financial Services Committee, Bessent discussed various economic concerns, with AI becoming a focal point. When questioned about the advancements in AI, particularly towards artificial general intelligence, he noted that Anthropic’s Mythos model was a significant leap forward. He stressed that the U.S. must prioritize the development of open-source AI models.

“We really need to cultivate more open-source models in the U.S.,” Bessent remarked. “If we allow large labs to dominate without regulation, it stifles innovation.” This statement underlines the need for balancing development oversight with the freedom to innovate.

Open-source AI models provide the public access to their underlying code and training processes, unlike closed-source models that are limited to select developers. Interestingly, some AI labs in China have reportedly learned from U.S. tech companies by using their closed-source models through a process called distillation, creating effective models at potentially lower costs.

As an illustration, a Chinese-developed open model was involved in addressing an incident where an OpenAI model breached containment and caused issues on another platform.

Bessent further explained, “One way to counter China’s advances is through our own model development. They essentially take U.S. models—what I would describe as a polite term for theft—and if we invest in our own open-source initiatives, it might deter users from opting for Chinese models.” He pointed out that many Chinese models appear to emulate existing U.S. models like Mythos and Claude.

In addition, U.S. tech companies have raised concerns about potential governmental restrictions on open-weight AI models, believing that such measures could hamper innovation, negatively impact cybersecurity, and diminish the competitiveness of the American AI sector.

A joint letter from several prominent tech companies—Nvidia, Microsoft, Meta, and others—was released in July advocating against broad restrictions on open model development. They argued for addressing illegal distillation through specific legal measures rather than imposing sweeping restrictions that could hinder important innovations in AI.

In August, the Trump administration introduced a voluntary AI framework that exempts open-source and open-weight models from pre-release government security reviews, focusing instead on proprietary models.