On 12 June 2026, Anthropic announced that the US government had ordered it to suspend access to Fable 5 and Mythos 5 for any foreign national, “whether inside or outside the United States”, including Anthropic’s own foreign-national employees. The company said the practical effect was that it had to disable both models for all customers while it worked out compliance. Access to other Anthropic models was not affected, but the principle had been established: the most advanced AI systems could be treated not as ordinary commercial software, but as strategic assets controlled by citizenship.
That would be striking under any circumstances. It becomes much more significant when read alongside Anthropic’s own argument about recursive self-improvement. In a separate Institute piece, Anthropic says AI is already accelerating AI development itself. Claude is no longer merely suggesting snippets of code for humans to paste into editors. It is writing and editing files, running code, debugging systems, carrying out experiments and increasingly taking over the practical middle layer of technical work. Anthropic says that as of May 2026, more than 80% of the code merged into its own codebase was authored by Claude, and that in Q2 2026 its typical engineer was merging eight times as much code per day as in 2024.
If those claims are broadly right, the US restriction is not merely a restriction on access to a product. It is a restriction on access to the multiplier.
That is the central point. If Mythos-class systems are merely impressive chatbots, then blocking access is commercially disruptive and politically provocative, but still comprehensible as an export-control decision. If, however, such systems are becoming the normal way in which most software development – including AI research – is conducted, then the policy touches the productive base itself. It does not just say who may use an American model. It says who may work at full speed inside the emerging technological order.
This is where the passport logic becomes dangerous for America. Silicon Valley has never been American in the narrow ethnic or citizenship sense. It has been American because the United States managed to gather capital, universities, companies and global talent in one place. The old bargain was simple: come here, build here, and the tools of the future will be available to you. A citizenship-based restriction on the most powerful AI systems weakens that bargain at precisely the moment when those systems may be becoming the means by which the next generation is built.
A Danish, Indian, French, Korean or Brazilian engineer in California who cannot use the same AI tools as an American colleague is not merely inconvenienced. If Anthropic’s multiplier is real and rising, that engineer is being forced to work at a lower speed. A fourfold or eightfold productivity difference is not a minor administrative matter. It is a career structure. It determines who can contribute, who can lead, who can remain relevant and where serious work can be done.
No multinational software company can live comfortably with that division for long. Modern engineering is collaborative. Teams are distributed across borders, passports and time zones. Codebases do not divide themselves neatly into citizen-safe and foreigner-safe compartments, unless one wishes to turn software development into airport security with pull requests. A company whose best tools are available only to part of its staff will eventually look for tools the whole team can use.
That is the gift to Europe, and perhaps especially to Mistral and any other serious non-American AI effort. Europe does not need to produce a perfect rival overnight. It needs a model good enough that European companies, universities and governments can build without asking Washington which passport is allowed to touch the future. A slightly weaker model that everyone on the team may legally use can be more useful than a stronger model wrapped in national-security restrictions. In the history of technology, the best tool does not always win. Very often, the tool people are allowed to use wins.
For years, European digital sovereignty has sounded worthy but vague, one of those phrases that emerges from policy seminars with its hair neatly combed and no obvious connection to plumbing. But this is plumbing. If advanced AI becomes the infrastructure of knowledge work, Europe cannot simply rent it from a foreign state that may restrict access by citizenship. Sovereignty ceases to be rhetoric when dependence becomes operationally visible.
This does not mean the American fear is irrational. If Anthropic are right about recursive self-improvement, then governments have reason to be frightened. A system that helps build better AI is not just another app. It is a machine for accelerating technological capacity. Such systems could improve science, medicine, engineering and public administration. They could also improve cyberattacks, surveillance, manipulation and weapon development. The multiplier is morally neutral in the most alarming possible way. It strengthens whatever intention controls it.
But a blunt citizenship restriction may protect the lead in the short term while damaging the ecosystem that created it. The United States became dominant partly because it could import talent and make that talent productive. If access to the frontier multiplier depends on citizenship, much of that talent will go elsewhere, and some companies will follow, while others will lose ground to competitors abroad. There may be no single exodus, no cinematic procession of engineers carrying laptops across the Atlantic. More likely, the shift will happen through hiring decisions, research partnerships, procurement rules, cloud contracts and start-ups quietly choosing to build around non-US stacks.
Once those ecosystems form, they will not necessarily return. Toolchains are sticky. Institutions are sticky. Procurement is sticky. Human beings, for all their alleged intelligence, are often just barnacles with budget meetings.
The deeper question is not whether Fable 5 and Mythos 5 return next week. They may. Anthropic and the US government may find a compromise, with licences, trusted-user categories, allied-nation carve-outs or some other bureaucratic contraption with a reassuring acronym. The deeper question is whether the precedent has now been set: frontier AI is strategic, access may be political and the multiplier may have a passport.
If that is the new world, then the coming AI race is not only about who has the best model. It is about who can build an ecosystem in which enough people are legally able to use it. The United States may still have the strongest models, the deepest capital markets and the largest chip capacity. But if the future is built by multinational teams using AI systems to create the next generation of AI systems, then exclusion becomes costly.
If the United States restricts access by citizenship, while the European Union restricts it only by residency or trusted legal presence, the latter will gain the stronger AI ecosystem over time.
The Industrial Revolution multiplied muscle. The computer revolution multiplied calculation. The AI revolution multiplies intention. If Anthropic are right, that multiplier is beginning to apply to itself. The future may not be built simply by the country with the most powerful model. It may be built by the people allowed to use the model that helps build the next one.