The Trump administration has taken new steps to control which companies and entities are allowed access to the most advanced AI models from Anthropic and OpenAI, a decision that until now rested with the labs themselves, two people familiar with the matter told CNBC.
Anthropic has controlled access to its Mythos cybersecurity model through a partner program called Project Glasswing, while OpenAI runs a similar consortium called Daybreak for its own cybersecurity-focused model. Going forward, both partner lists will require explicit government approval, according to CNBC’s reporting. The shift follows the launch this week of a White House initiative called Gold Eagle, described as an AI clearinghouse for cyber vulnerabilities, which sources said will put the administration in a position to greenlight which companies can access new frontier models.
A White House official told CNBC the government does not “provide approvals for AI releases” from private companies, and said any engagement, testing or meetings between government experts and AI labs is voluntary, with decisions on release timing and scope remaining 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 said, pointing to President Trump’s executive order on AI security. CNBC’s sources disputed that characterization of the arrangement as voluntary.
The framework traces to Executive Order 14409, signed June 2, which asks frontier AI developers to give the federal government access to covered models for up to 30 days before releasing them to other trusted partners. The order states it should not be read as creating a mandatory licensing or preclearance requirement. In practice, the government has already used the arrangement to shape two major releases. Last month, the administration blocked Anthropic’s Claude Mythos 5 and Fable 5 models over national security concerns, reinstating access after weeks of negotiation. OpenAI separately said it would limit its GPT-5.6 release to “trusted partners” to comply with a government request to stagger the rollout.
The tightening access comes at a politically uncomfortable moment for the administration’s stated goal of maintaining U.S. leadership in AI. Chinese startup Moonshot AI launched its Kimi K3 model the same day CNBC’s report published, with the company saying it matched or exceeded the performance of Fable 5 and GPT-5.6 on at least one independent benchmark. David Sacks, a former White House AI adviser, criticized the gating approach, warning in a social media post that bogging down U.S. releases risked ceding ground to competitors operating under different rules.
Separately, Bloomberg has reported that Treasury Secretary Scott Bessent helped develop an internal proposal for a standing, independent AI oversight body reporting to the Securities and Exchange Commission, modeled on the finance industry’s FINRA, that would screen frontier models for dangerous capabilities before release. The proposal remains under internal review and has not been formally adopted.
The White House has not said how long the current partner-approval arrangement will remain in place or whether it will be formalized through additional executive action.
