President Donald Trump points his finger as he signs an executive order on AI next to Senate Commerce Committee Chairman Ted Cruz, R-Texas, left, and Commerce Secretary Howard Lutnick, in the Oval Office at the White House in Washington, Dec. 11, 2025.
Al Drago Reuters
The Trump administration has taken new steps to ensure more control over the release of artificial intelligence models in the future by placing restrictions on which companies and organizations are allowed access to the latest models, two people familiar with the matter told CNBC.
Until now, that decision was in the hands of America’s AI giants.
Both Anthropic again OpenAI they have decided which companies and agencies have access to their most powerful models, and they often include large corporate clients.
Anthropic introduced its Mythos high-performance cybersecurity model to several partners through Project Glasswing. OpenAI was asked by management to release its latest GPT-5.6 release, and it has a similar organization called Daybreak for its cybersecurity model.
A White House official told CNBC that he doesn’t give approvals for AI outsourcing to private companies.
The official said any interviews, inspections or meetings with government experts are “voluntary” and that “decisions about the timing and scope of the releases are entirely up to the companies,” referring CNBC to Trump’s latest order.
“The administration continues to work with all U.S. border labs to strengthen the security of these technologies without stifling innovation,” they wrote.
However, last month the Trump administration blocked Claude Mythos 5 and Fable 5 due to “national security concerns,” restoring access after weeks of intense negotiations with Anthropic. OpenAI last month said it would limit new AI models to “trusted partners” to comply with government requests.
The White House is walking the line at a time when sophisticated AI tools pose a major threat to cyber security and cheap, lightweight models from China are quickly closing the gap with America’s border labs.
Chinese startup Moonshot AI launched its Kimi K3 model on Friday, which took a big hit in Fable and GPT-5.6 performance, even surpassing US flagship models in at least one independent benchmark.
David Sacks, founder of Craft Ventures and former White House AI director, called Kimi’s success “remarkable.”
“This is how you lose the AI race,” he wrote. “The whole world will not play by our rules if we ignore them.”
The administration has already taken several steps to reshape AI oversight in recent months, starting with President Donald Trump’s June order, which asked companies to voluntarily give the government early access to models to be tested.
This week, the administration launched its own program, called “Golden Eagle,” which aims to work with private companies to find and fix cyber vulnerabilities.
The so-called clearinghouse will put the White House in control of green-lighting companies that cannot access new AI models, according to a person familiar with the matter, who spoke on condition of anonymity to discuss non-public information.
The administration’s moves have left the future of company-led initiatives like Project Glasswing and OpenAI’s Daybreak in doubt.
Going forward, according to one person, this rollout will require express government approval in which partners can be excluded.
— CNBC’s Megan Cassella and Ashley Capoot contributed reporting.
