White House assumes de facto control over frontier AI access

The Trump administration has taken effective control of which organisations can access the most advanced AI models developed by OpenAI and Anthropic. A new White House programme called Gold Eagle now requires explicit government approval for the partner lists of these US labs, stripping the companies of their previous authority to decide who uses their technology.

Until now, Anthropic granted access to its Mythos cybersecurity model through an internal initiative called Project Glasswing. OpenAI managed its own cyber model access via a programme named Daybreak. The White House officially maintains that company participation is "voluntary" and that the government does not "provide approvals for AI releases."

However, the operational reality contradicts this stance. Last month, the administration blocked Anthropic’s Claude Mythos 5 and Fable 5 over national security concerns, only reinstating access after weeks of negotiations. In June, OpenAI stated it would limit new models to "trusted partners" specifically to comply with government requests.

Gold Eagle, launched this week as a clearinghouse for AI cyber vulnerabilities, operationalises a June executive order into a concrete gating mechanism. Without any new legislation or formal regulatory agency, the US government has acquired de facto distribution authority over frontier artificial intelligence.

For European businesses and investors, this represents a sudden and opaque barrier to critical infrastructure. European tech firms, cybersecurity agencies, and enterprises integrating these frontier models into their operations now face a political veto rather than a commercial one. Access to the underlying drivers of productivity and security is no longer determined by a company's technical merit or ability to pay, but by an unlegislated White House programme.

This tightening of access comes at a geopolitically awkward moment. On the same day Gold Eagle launched, Chinese lab Moonshot AI released its Kimi K3 model. According to at least one independent benchmark, Kimi K3 matched or exceeded the capabilities of both Fable and OpenAI’s GPT-5.6.

The administration is attempting to secure frontier AI against Chinese exploitation, but the move risks ceding ground. Former White House AI czar David Sacks summed up the competitive anxiety: “This is how you lose the AI race. The rest of the world won’t play by our rules if we bog ourselves down.” For Europe, the immediate risk is being caught between an increasingly protectionist Washington and a rapidly closing capability gap in Beijing.