TL;DR
The Trump administration is dictating access to border AI models, shifting that decision from Anthropic and OpenAI to the government through the Golden Eagle program.
The Trump administration is now dictating which companies and institutions get access to frontier artificial intelligence models from Anthropic and OpenAI, CNBC reported Friday, citing two people familiar with the matter. Until now, laboratories made this decision themselves. Anthropic controlled access to the Mythos cybersecurity model through an initiative called Project Glasswing. OpenAI ran a similar program called Daybreak for its cyber model. In the future, these partner lists will require explicit government approval.
White House official told CNBC the government doesn’t do that”Provide approvals for AI releases” and company involvement “voluntary.“However, the administration blocked Anthropic’s Claude Mythos 5 and Fable 5 last month over national security concerns and restored access only after weeks of negotiations. OpenAI said in June it would restrict new models.”reliable partners” to meet government demands.The gap between official position and operational reality is the story. The White House presented the Golden Eagle this weekan AI clearinghouse for cyber vulnerabilities, and according to CNBC’s source, the program will be responsible for the White House greenlighting companies to access new AI models.
The timing is politically unpalatable. Moonshot AI’s Kimi K3 launched that day and matched or surpassed the Fable and GPT-5.6 in at least one independent benchmark. Former White House AI czar David Sacks called it “in connection with” and wrote: “So you lose the AI race. If we bog ourselves down, the rest of the world won’t play by our rules.“The administration is trying to protect border AI against Chinese exploitation while simultaneously watching Chinese labs close the capability gap in real time.
The change is structural, not temporary. Trump’s June executive order required AI companies to give the government early access to models for testingvoluntarily framed. Golden Eagle performs operations that require something closer to a gate mechanism. If Anthropic and OpenAI can’t release the most capable models on their list of partners without government approval, then the U.S. government has gained de facto distribution power over border AI without legislation, without a regulatory body, and through a program the White House insists is optional.
