US balances AI innovation and safety in tech race with China

US Treasury Secretary Scott Bessent told the House Financial Services Committee on Tuesday that the Trump administration has spent six months developing a framework to accelerate artificial intelligence innovation while ensuring safety safeguards amid competition with Beijing. Speaking during the panel's annual hearing on the international financial system, Bessent said the administration aims to maintain what he called "the optimal growth rate" by addressing both innovation and security concerns simultaneously.

Bessent warned against allowing major technology companies to dominate regulatory discussions, arguing that such capture would stifle competition. "We can't let these large labs have regulatory capture, because that will stop innovation," he said, advocating instead for the expansion of open-source and open AI models to democratize access to the technology.

Competition with China

The Treasury secretary framed the AI race in adversarial terms, accusing Chinese developers of appropriating American intellectual property to advance their own systems. "The Chinese models, they distill from the US models, which is a polite scientific word for steal," Bessent said, adding that restricting access to advanced semiconductors remains a critical component of Washington's strategy. "AI runs on the advanced chips that the US has a substantial lead in, so making sure that other countries do not have access to those advanced chips is very important," he noted.

Bessent stressed that leadership in AI development carries profound implications for global security, dismissing concerns that capability alone should dictate policy. "It does matter whether the good guys or the bad guys have this," he said. "As the person who manages the relationship with China on AI, it matters a great deal."

Financial sector resilience

Bessent said Treasury began engaging with major financial institutions in March to assess AI risks within the banking sector before expanding those consultations to smaller institutions over subsequent months. These efforts feed into Project Gold Eagle, which he described as "a clearinghouse for everything that we learn about resiliency" and a forum for sharing cybersecurity information across federal agencies that have examined AI models for more than five months.