White House Asserts Control Over Frontier AI Model Access, Sidelining Tech Giants

The Trump administration has moved to take direct control over which companies and entities may access the most advanced artificial intelligence models released by American AI developers, a shift that curtails authority previously held exclusively by firms like Anthropic and OpenAI.

Two sources familiar with the matter told reporters that the administration is now dictating access to frontier AI models — decisions that had until recently rested entirely with the companies themselves.

Both Anthropic and OpenAI had established their own controlled rollout programs. Anthropic unveiled its most capable Mythos cybersecurity model to a select group of partners through an initiative called Project Glasswing. OpenAI operates a comparable consortium known as Daybreak for its own cybersecurity model and was asked by the administration to gate its recent GPT-5.6 release.

The administration's new program, dubbed "Gold Eagle," launched this week as a clearinghouse intended to identify and remediate cyber vulnerabilities in collaboration with the private sector. According to a source familiar with the matter who spoke on condition of anonymity, the clearinghouse would give the White House authority to greenlight which companies can access new AI models going forward.

A White House official pushed back on characterizations of a formal approval regime, telling reporters that the administration does not provide approvals for AI releases from private companies and that any engagements with government experts are "voluntary." "Decisions on timing and scope of releases rest entirely with the companies," the official said, pointing to President Donald Trump's recent executive order.

That executive order, issued in June, asked companies to voluntarily grant the government early access to models for testing purposes. But the administration's actions have at times gone further. Last month, it blocked Anthropic's Claude Mythos 5 and Fable 5 over what it described as "national security concerns," restoring access only after weeks of negotiations with the company. OpenAI separately said last month it would limit new model access to "trusted partners" in response to government requests.

The moves have cast uncertainty over the future of company-led programs like Project Glasswing and Daybreak. According to one source, future rollouts will require explicit government approval for which partners may participate.

The administration's expanding oversight comes as Chinese AI developers continue to close the performance gap with their American counterparts. Chinese startup Moonshot AI on Friday unveiled its Kimi K3 model, which, according to at least one independent benchmark, matched and in some cases outperformed both Fable and GPT-5.6.

David Sacks, founder of Craft Ventures and Trump's former White House AI czar, characterized the development as a warning sign. "This is how you lose the AI race," Sacks wrote. "The rest of the world won't play by our rules if we bog ourselves down."

The White House is attempting to balance two competing pressures: reducing the cybersecurity risks posed by sophisticated AI tools while avoiding regulatory friction that could slow American AI development relative to China. How the administration navigates that tension — and whether company-led access programs survive in any meaningful form — will likely determine the pace and shape of frontier AI deployment in the months ahead.