The landscape of artificial intelligence governance in the United States has undergone a profound shift. According to a recent report by CNBC, the Trump administration is now directly controlling which organizations and entities can access the most advanced frontier AI models from leading labs like Anthropic and OpenAI. Until now, these decisions were made internally by the companies themselves. Anthropic managed access to its cutting-edge Mythos cybersecurity model through a program called Project Glasswing, while OpenAI operated a similar initiative named Daybreak for its cyber models. Going forward, any partner list associated with these programs must receive explicit approval from the White House.
This represents a dramatic change in the relationship between private AI developers and the federal government. The administration's official stance, as conveyed by a White House official to CNBC, is that the government does not "provide approvals for AI releases" and that company participation is "voluntary." However, operational reality tells a different story. Last month, the administration blocked the release of Anthropic's Claude Mythos 5 and Fable 5 models over national security concerns. Access was only reinstated after weeks of intense negotiations. Such actions demonstrate that while the process may be framed as voluntary, the consequences of non-compliance are severe.
The Gold Eagle Program
This week, the White House officially launched the Gold Eagle program, an AI clearinghouse designed to coordinate the handling of cyber vulnerabilities. According to sources cited by CNBC, Gold Eagle will place the White House in charge of greenlighting which companies can access new AI models. This structure effectively gives the government de facto distribution authority over frontier AI. The program operates without explicit legislation or a dedicated regulatory agency, relying instead on executive orders and voluntary compliance. The Trump administration's June executive order requested that AI companies provide early access to models for testing, but Gold Eagle transforms that request into a gating mechanism.
The implications are enormous. If Anthropic and OpenAI cannot release their most capable models without government approval of the partner list, then the US government has acquired a level of control over AI dissemination that rivals that of any regulatory body. This shift is particularly significant given the speed of AI development and the intense global competition in this field.
National Security Versus Innovation
The timing of this policy shift could not be more politically sensitive. On the same day that Gold Eagle was announced, Chinese AI lab Moonshot AI released its Kimi K3 model. Independent benchmarks indicate that Kimi K3 matches or exceeds the performance of Anthropic's Fable and OpenAI's GPT-5.6, at least on one key evaluation. This development underscores the rapid pace at which Chinese labs are closing the capability gap with their American counterparts.
David Sacks, former White House AI czar, expressed concern about the situation. Writing on social media, he noted: "This is how you lose the AI race. The rest of the world won't play by our rules if we bog ourselves down." Sacks' comment highlights a central tension: the administration's efforts to secure frontier AI against Chinese exploitation may inadvertently hinder the very innovation that keeps the US ahead in the global AI race.
The administration's approach is a double-edged sword. On one hand, it aims to prevent sensitive AI technology from falling into the hands of adversaries who might use it for malicious purposes, such as cyberattacks, disinformation campaigns, or military applications. On the other hand, restrictive access policies could slow down research, reduce the pace of breakthroughs, and push talent and investment to jurisdictions with more favorable regulatory environments.
Historical Context and Regulatory Landscape
To understand the significance of this shift, it is helpful to consider the broader history of AI governance in the United States. Prior to the Trump administration, the federal government largely took a hands-off approach to AI regulation. The Obama administration released a report on AI in 2016 that focused on ethical principles but did not impose controls. The Biden administration issued an executive order on AI safety in October 2023, which introduced testing requirements for powerful models, but still left access decisions largely to the companies.
Trump’s approach has been more interventionist. In June 2024, an executive order mandated that companies developing frontier AI models must share safety test results with the government. This was followed by the creation of the AI Safety Institute, which was intended to evaluate models but not to control access. The Gold Eagle program goes further by inserting the White House directly into the distribution chain.
Critics argue that this move creates a dangerous precedent. Without a clear legal framework, the administration can arbitrarily block or approve access to AI models, potentially based on political considerations rather than objective security assessments. Supporters counter that the government needs robust mechanisms to protect national security in an era when AI capabilities are advancing faster than regulation.
Impact on the AI Ecosystem
The consequences of this policy are already being felt. OpenAI stated in June that it would limit new models to "trusted partners" to comply with government requests. This language suggests that the company is preemptively restricting access to avoid conflict with the administration. Anthropic has not publicly commented on the specifics of the Mythos 5 and Fable 5 blockages, but industry insiders report that the episode has caused significant internal debate about the viability of continuing to develop and release frontier models under such conditions.
Smaller AI startups are also affected. Many rely on access to foundation models from larger labs to build their own applications. If the White House restricts which companies can obtain models, it could create an uneven playing field, favoring well-connected firms over innovative newcomers. This could stifle competition and concentrate AI power even further among a handful of incumbents.
International partners are watching closely. The European Union has already implemented its AI Act, which imposes strict requirements on high-risk AI systems. The US approach, by contrast, is ad hoc and executive-driven. This lack of predictability could deter foreign companies from collaborating with American AI labs, fearing that access may be cut off without warning.
Technical and Ethical Dimensions
From a technical standpoint, the concept of "frontier AI models" is itself somewhat fluid. These are generally defined as the most advanced and capable models that push the boundaries of what AI can do. However, as models improve, the frontier moves. What is considered frontier today may be commodity tomorrow. This makes it difficult to craft stable policies around access controls.
Ethically, the centralized control of AI raises questions about accountability. If the White House approves a partner list and that partner uses the model to cause harm, who is responsible? The government? The lab? The partner? The current framework does not address such scenarios. Additionally, there is a risk of politicization: a future administration could use the same mechanism to block models for reasons unrelated to national security, such as protecting incumbents from competition or suppressing certain types of speech.
The AI research community has largely reacted with alarm. Many researchers argue that open access to models is essential for scientific progress and for the development of safety measures. If only a select few companies and government-approved entities have access to the most capable models, then independent researchers cannot scrutinize them for flaws or biases. This could lead to a situation where dangerous vulnerabilities go undetected until it is too late.
Global Reactions and Comparisons
China has responded to the US shift by doubling down on its own AI development. The release of Moonshot AI’s Kimi K3 is just one example. Chinese government funding for AI research has increased significantly, and the country has established its own set of AI regulations that prioritize state oversight and control. In a way, the US and Chinese approaches are converging: both governments seek to manage AI risks through state intervention, though the rhetoric differs.
The European Union's AI Act takes a different path, focusing on harmonized rules across member states. It does not give a single authority the power to block model releases. Instead, it requires conformity assessments and transparency. This regulatory model may become more attractive to global companies looking for predictability.
Other countries, such as the United Kingdom and Japan, have signaled they will take a more hands-off approach, hoping to attract AI innovation. The divergence in regulatory strategies could lead to a fragmentation of the global AI ecosystem, where companies choose jurisdictions based on the ease of doing business.
Long-Term Implications
The structural change represented by Gold Eagle is not a temporary measure. Even if a future administration reverses it, the precedent of government control over AI access will remain. The program operationalizes the executive order into a lasting mechanism that could be expanded or repurposed. Moreover, it establishes a norm that the government has a legitimate role in deciding who can use the most powerful AI tools.
For the United States to maintain its leadership in AI, it must find a balance between security and openness. The current approach risks tipping too far toward control, which could inhibit the very innovation that gives the US its edge. Meanwhile, Chinese labs continue to advance, unencumbered by the same restrictions. The race is not yet lost, but the margin for error is shrinking.
The White House insists that the Gold Eagle program is voluntary and that it does not provide approvals. Yet the actions taken against Anthropic speak louder than words. The gap between official policy and operational practice is where the real story lies. As the AI landscape evolves, the debate over who controls access to the most powerful models will only intensify, with profound implications for national security, economic competitiveness, and the future of technology itself.