The Trump administration’s embargo on Anthropic’s latest models scorches its hopes to export American AI

One goal of America’s AI Action Plan, the Trump administration’s artificial intelligence (AI) strategy announced last year, is to “drive adoption of American AI systems, computing hardware, and standards throughout the world.” To execute this plan, the administration developed a program to promote exports of America’s “full AI technology stack.” Now, it has set fire to its own strategy by telling Anthropic to deny foreign nationals access to its Mythos 5 and Fable 5 models.

This directive has ignited a raging conflagration in numerous parts of the world where ambitions for “AI sovereignty” have been smoldering. Governments have sound reasons to want agency over the technology used within their borders, such as supply chain resilience, national security, and competitiveness. But a leading factor has been distrust of U.S. technology, U.S. companies, and the U.S. government. Coming alongside the disruption of oil markets and other slights, the Anthropic cutoff adds fuel to these aspirations.

Europe has long bemoaned its shortage of technology leaders and sought to wean itself from dependence on American companies by building up homegrown competitors. Even before the model cutoff, I heard frustration from foreign diplomats that Project Glasswing, the pre-release review of the Mythos model, was initially restricted to U.S. participants. Henna Virkkunen, the EU executive vice president in charge of tech sovereignty, called the cutoff “another wake-up call … for Europe.” The EU is not alone: As of 2026, some 46 countries are investing government funds in some 135 sovereign AI projects. As a Fortune headline aptly described, the cutoff has sparked “a global scramble for sovereign AI.”

My déjà vu

I’ve seen this kind of firestorm before. I was acting secretary of commerce when news of Edward Snowden’s leaked files hit. When the story broke about the PRISM program’s intelligence gathering on international email traffic, it showed a National Security Agency (NSA) slide with the logos of several U.S. internet companies. The report incorrectly claimed that the NSA was “tapping directly into the central servers” of these companies—feeding suspicions that U.S. intelligence was monitoring the entire firehose of transatlantic traffic.

I knew this report would trigger a catastrophe for American companies in Europe and elsewhere. The next day, President Barack Obama inadvertently magnified the impact by reassuring his domestic audience that the PRISM surveillance “does not apply to U.S. citizens.” This raised a red flag for people everywhere else, leading to a search for alternatives to American technology, including data centers located in Europe, European service providers, and then-Chancellor of Germany Angela Merkel’s call for a “European internet.”

In the weeks that followed, I ruffled some feathers at the White House by pushing to inject the commercial impact into a response discussion focused entirely on national security. In time, that damage sunk in, and the Obama administration made sweeping changes to the transparency of U.S. intelligence collection by declassifying a wide body of information, stepping up diplomacy, and appointing an independent board of experts for a broad review of signals intelligence collection. Affected companies made their postures clearer and issued “transparency reports” clarifying the modest volume of intelligence requests they actually received. In 2014, the White House issued a presidential policy directive that provided additional safeguards for foreign intelligence surveillance and began a multinational discussion of norms for government access to information.

Even so, distrust endured. The Court of Justice of the European Union has twice declared U.S.-EU data transfer arrangements invalid under European law, requiring new frameworks and mechanisms. The current framework is under review in that court yet again. It’s not just about fears of government surveillance; the impacts of social media and other platforms and their concentration of power added to the antipathy toward American tech companies and fueled an expanding digital agenda that includes a “EuroStack.”

Even as a short-term measure, the Mythos-Fable embargo adds fuel to this long-running fire, and that fire has become global. Like the handling of the Snowden leaks, it brands allies as adversaries, moving them to do the same.

Repairing the damage

The Trump administration needs to deal with global aspirations for AI sovereignty better than it did in reacting to the release of the Mythos and Fable models. The AI action plan calls for leveraging America’s AI advantages “into an enduring global alliance.” Doing so becomes a challenge when you keep fracturing alliances. So does exporting the full American AI stack if you refuse to provide the most advanced models to any foreign nationals.

The administration should follow through on its plans for diplomacy to promote AI exports and its “Pax Silica.” At India’s AI Impact Summit in February, White House Office of Science and Technology Policy Director Michael Kratsios offered to partner with other countries to help them achieve strategic autonomy. In a similar vein, the Pax Silica declaration brought together 24 like-minded governments to cooperate on resilient and secure AI supply chains, including “efforts to partner on strategic stacks.” This tone is consistent with a report that colleagues at the Brookings Institution, the Centre for European Policy Studies, and I issued at that time. The report laid out dependencies across the AI stack that make it impossible for any nation—including the U.S. and China—to go it entirely alone in developing AI, along with the strategies of managed interdependence needed to achieve sovereignty where it is beneficial and feasible. Rather than blindsiding allies and key trade partners, the administration should reinvigorate collaboration with trusted partners in the G7 and AI safety reviews through the international network of safety institutes, including for leading-edge models.

Trust is essential currency in the digital economy, and a more coherent and complete U.S. AI policy is essential to building trust in American AI. In a few head-spinning weeks, the administration has gone from adamantly opposing any regulation on AI—to the point of supporting a complete moratorium on state laws—to voluntary review of cutting-edge models, negotiated down from a 90-day “voluntary” review period to 30 days, to effectively shutting down the most advanced models.

In place of haphazard lurches, it is time for Congress to enact meaningful AI regulation. Polls by Gallup, Pew, and Johns Hopkins show extraordinary majorities of Americans are worried about AI and support regulation of safety, security, and other aspects of the technology.

A structure to rationalize review of foundational models—along with resources to support such work—is a necessary first step. But adequate regulation of AI will need to include a broader layer of safety and security regulation. This must encompass reasonable care in developing and deploying AI systems that may have a material impact on people and society, such as risk assessments, monitoring, and organizational accountability.

Fears that such requirements could hinder AI development and adoption are overblown; a legion of firms and business organizations have focused on AI governance at the organizational level and developed best practices that demonstrate standards of care. Privacy is another vital step toward protecting individuals in an AI-driven world where their personal information may be identified in training data and they make growing parts of their lives available to AI agents. AI regulation without privacy protections would be like a body without an immune system.

As the Snowden experience demonstrated, transparency is essential to trust; it is also needed to build greater understanding of AI for policymakers and the public. Thus, baseline transparency requirements should also be part of AI regulation.

The premise of the AI action plan is that American companies have developed the most innovative and powerful AI. To sustain this position, the American government needs to enact laws and governance that also make American AI the most trustworthy. The nation’s export strategy cannot succeed without trust from governments and people around the world.