White House expands role in who can test top AI models

The Trump administration is asserting more influence over who gets early access to the newest artificial intelligence systems, CNBC reported, citing two people familiar with the matter. For investors tracking AI, the shift is relevant because access rules can affect product launches, enterprise sales and how companies manage cyber risks tied to powerful models.

A frontier AI model is a company’s most advanced system, usually capable of handling complex tasks across coding, writing, research and cybersecurity. CNBC reported that decisions over who could test or use those models had largely sat with U.S. AI companies, including Anthropic and OpenAI.

Anthropic previously introduced its Mythos cybersecurity model to a limited group of partners through an initiative called Project Glasswing, according to CNBC. OpenAI, meanwhile, was asked by the administration to restrict access to its GPT-5.6 release and has a comparable cybersecurity consortium called Daybreak, CNBC reported.

White House says company releases remain voluntary

A White House official told CNBC that the administration does not approve AI releases by private companies. The official said any work with government experts, including testing or meetings, is voluntary and that decisions about release timing and scope remain with the companies.

“The Administration continues to collaborate with all of America’s frontier labs to strengthen the security of this technology without stifling innovation,” the official told CNBC, pointing to President Donald Trump’s recent executive order.

That June order asked AI companies to voluntarily give the government early access to models for testing, CNBC reported. The administration has since taken additional steps that put Washington closer to the release process.

Last month, the Trump administration blocked Anthropic’s Claude Mythos 5 and Fable 5 over “national security concerns,” CNBC reported. Access was restored after weeks of negotiations with Anthropic. OpenAI also said last month that it would limit new AI models to “trusted partners” in order to comply with government requests, according to CNBC.

Gold Eagle adds a government-run clearinghouse

This week, the administration launched Gold Eagle, a program designed to work with the private sector on finding and fixing cybersecurity vulnerabilities, CNBC reported. A person familiar with the matter told CNBC the clearinghouse would put the White House in charge of deciding which companies can access new AI models.

Another person familiar with the matter told CNBC that future rollouts would require explicit government approval for participating partners. CNBC reported that the development leaves the future of company-led access programs such as Project Glasswing and Daybreak uncertain.

The policy push comes as U.S. officials weigh security concerns against competition from overseas AI developers. CNBC reported that Chinese startup Moonshot AI unveiled Kimi K3 on Friday, a model it said had largely caught up with Fable and GPT-5.6 and surpassed U.S. frontier models on at least one independent benchmark.

David Sacks, founder of Craft Ventures and the former White House AI czar, called the Kimi development “concerning” in a post on X cited by CNBC. “This is how you lose the AI race,” he wrote. “The rest of the world won’t play by our rules if we bog ourselves down.”

This story draws on original reporting from CNBC.