Austria Urges the EU to Host Anthropic After U.S. Locks Foreigners Out of Its Top Models

When Washington pulled the plug on foreign access to Anthropic's most capable models, the loudest response did not come from Brussels, Paris or Berlin. It came from Vienna, in the form of a single letter with an audacious ask: bring the company itself to Europe.

In a letter to European Commission Executive Vice President Henna Virkkunen, released by the Austrian government on June 28, Austria's State Secretary for Digitalization Alexander Pröll urged the EU to consider hosting Anthropic inside the bloc's borders. The pitch was not for another data center or a procurement contract. It was, in effect, an invitation to relocate one of America's two leading AI labs to a jurisdiction where Europeans could not be locked out of it.

"Let us jointly explore the strategic establishment and participation of Anthropic within the European Union," Pröll wrote, according to text released by his office. "With legal certainty, market access, capital and a set of values that suits this company."

He did not spell out how that would actually happen, and he conceded the idea would draw skepticism. But he framed the stakes in stark, almost philosophical terms. "The real question is not whether it is easy," Pröll wrote. "The question is whether we Europeans are prepared to be the architects of our technological future, or whether we wish to remain mere administrators of decisions made elsewhere."

Anthropic did not immediately respond to a request for comment on the Austrian proposal.

The trigger: a June 12 export-control directive

The letter is a direct reaction to an abrupt shift in U.S. policy. On June 12, the U.S. issued export-control guidance ordering a complete block on all foreign nationals' access to two of Anthropic's frontier systems, Claude Mythos 5 and Claude Fable 5, on national-security grounds. Because Anthropic had no reliable way to verify the nationality of individual users, the practical effect was blunt: the company restricted access to both models for everyone, U.S. users included, rather than risk violating the directive.

Then came a partial walk-back. In a confidential letter dated June 26, the U.S. Commerce Department notified Anthropic that a defined set of roughly 100 American companies and institutions could resume using Mythos 5 — while warning that the approved list could change at any time. Foreign users remained shut out. A similar logic is surfacing elsewhere in the industry: when OpenAI unveiled its next-generation GPT-5.6, it reportedly made the model available first only to institutions designated by the U.S. government, rather than releasing it broadly as in the past.

For European policymakers, the sequence crystallized a fear that has been building for years. The most advanced AI is now gated not by price or compute, but by permission — and that permission is being handed out in Washington.

A continent already reaching for sovereignty

Pröll's letter did not land in a vacuum. Earlier in June, the European Commission unveiled a sweeping technological-sovereignty package built around the Cloud and AI Development Act (CADA), which aims to roughly triple Europe's data-center capacity within five to seven years and steer priority grid access toward facilities that use European-made chips. Alongside it sits a "Chips Act 2.0," targeting a doubling of the EU's share of global semiconductor production — currently under 10 percent — by 2030.

Virkkunen, the commissioner now receiving Austria's letter, has been the public face of that push, and she has not been shy about naming the underlying anxiety. "We don't want a situation where someone holds a so-called 'kill switch' over our services," she said at a briefing on the package, in remarks widely read as aimed at the United States. She has also been candid about the limits. "Since 80 percent of our technology comes from outside Europe, it is impossible to build capabilities in this field in a short period," she acknowledged, suggesting visible results may not arrive before 2030.

That tension — urgent political will, distant technical reality — is exactly where Austria's proposal sits.

Is "hosting" a U.S. lab even possible?

The honest answer is that nobody knows, and the obstacles are formidable. Anthropic is a U.S. company, its frontier models are trained and governed under U.S. export law, and the very directive that prompted Pröll's letter is precisely the instrument Washington would use to stop those models from being run abroad for foreign users. A subsidiary in Vienna or Dublin does not, on its own, escape U.S. jurisdiction over the underlying technology. Capital, talent and the latest accelerator chips also remain heavily concentrated on the American side of the Atlantic.

Pröll's own framing — that the point is not whether it is "easy" — reads less like a concrete relocation plan than a deliberate provocation. By asking the unthinkable out loud, Austria has reframed the sovereignty debate: the scarce resource in 2026 is no longer just compute or engineers, but access itself, and Europe is discovering it has very little control over the supply.

There is also a strategic logic to floating the idea even if it fails. A credible European overture gives the bloc leverage in any future negotiation over carve-outs to U.S. export controls, and it signals to American labs that an alternative home — one offering legal certainty and a large single market — is at least theoretically on the table.

What to watch

The immediate test is whether Virkkunen and the Commission engage with the proposal at all, or quietly let it drop as an Austrian solo act. The CADA and Chips Act 2.0 measures still face months of negotiation among member states and the European Parliament before passage, and Austria's letter will sharpen the debate over how far Europe is willing to go.

Watch, too, for any signal from Anthropic itself, which has so far said nothing, and for whether Washington expands or narrows its June directive as the approved-company list evolves. The deeper question Pröll posed will outlast this news cycle: in an era when frontier AI ships with a government-controlled access list attached, can any region that does not build its own models claim to be sovereign at all?