The White House launched a clearinghouse called Gold Eagle to decide which companies get access to frontier AI models, sidelining the labs that built them

Washington Built a Gate Around America's Best AI Models

For years, if you wanted early access to the most powerful AI systems from OpenAI or Anthropic, you asked the labs. Not anymore.

A new government clearinghouse called Gold Eagle, launched this week, puts the Trump administration in charge of greenlighting which companies receive access to new frontier AI models from the two labs 1. Anthropic had controlled access to its Mythos cybersecurity model through a partner program called Project Glasswing. OpenAI ran a similar consortium called Daybreak for its own cyber model. Going forward, those partner lists will require explicit government approval 2.

A White House official told CNBC that participation is voluntary and that companies retain full control over the timing and scope of releases 1. The administration's own executive order, signed June 2, says the program should not be understood as creating "a mandatory governmental licensing, preclearance or permitting requirement" for AI development 3.

The operational record tells a different story. Last month, the administration blocked Anthropic's Claude Mythos 5 and Fable 5 over national security concerns and reinstated access only after weeks of intense negotiations 1. OpenAI said in June it would limit new models to trusted partners to comply with government requests 2. The June executive order also seeks government access to new models for up to 30 days before public release, during which officials select which companies qualify as early-access partners 3.

So the government can preview a new model for a month, decide who gets to build on it, and has already demonstrated a willingness to shut releases down entirely.

The timing compounds the problem. On the same day CNBC reported on Gold Eagle, Chinese startup Moonshot AI released its Kimi K3 model, which largely matched the performance of Anthropic's Fable and OpenAI's GPT-5.6 and outperformed both in at least one independent benchmark 1.

David Sacks, founder of Craft Ventures and the former White House AI czar, called the Kimi release concerning. "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" 1.

The security case behind Gold Eagle is genuine. Frontier models with advanced cyber capabilities could be weaponized, which is why the executive order asks companies to participate in benchmarking to determine whether a model qualifies as a "covered frontier model" subject to access restrictions 3. Anthropic's Project Glasswing reportedly uncovered more than 10,000 cybersecurity vulnerabilities in its first month of operation 3. The administration is trying to keep powerful offensive tools out of the wrong hands, and that goal is not frivolous.

But the mechanism leaves a structural question that the White House has not addressed. If OpenAI and Anthropic cannot release their most capable models without government sign-off on who can use them, then the executive branch has acquired distribution authority over frontier AI. No legislation authorized it. No regulatory agency administers it. The program the White House describes as optional has already been used to block a major lab's releases for weeks.

A startup that once negotiated access directly with a lab now faces an additional approval layer. The labs themselves are watching their distribution control migrate to an institution that operates by executive order rather than statute.

Sacks helped design the AI policy this administration is now executing. His eight-word verdict on where the competitive landscape is heading is the sharpest warning yet from inside that tent, and it landed the same week the gate went up.

The frontier models built in America now have a federal gatekeeper.