By Stuart Kerr, Technology Correspondent, LiveAIWire
Austria became the first European Union member state to formally ask the European Commission to explore hosting Anthropic within EU borders — a direct response to a June 12 US export-control order that cut 450 million Europeans off from the most powerful AI models on the planet, without warning, without negotiation, and without appeal.
The letter, sent by Austrian State Secretary for Digitalization Alexander Pröll to European Commission Executive Vice President Henna Virkkunen on June 28, asked member states to jointly explore “the strategic establishment and participation of Anthropic within the European Union,” offering legal certainty, single-market access, and capital as incentives. Pröll did not say how such a relocation would work in practice. He acknowledged that skepticism about its feasibility was entirely reasonable. But the letter was not really about moving a company. It was about something else: who controls the switch.
What Happened on June 12 and Why It Shocked European Governments
On June 12, 2026, the US Commerce Department’s Bureau of Industry and Security, acting under Commerce Secretary Howard Lutnick, issued an export-control directive ordering Anthropic to suspend access to Fable 5 and Mythos 5 — its two most advanced AI models, for any foreign national, whether inside or outside the United States, including Anthropic’s own non-citizen employees.
The legal instrument invoked was the Export Control Reform Act, applied to a commercially available AI model for what appears to be the first time. The stated reason was national security. The trigger was contested: the US government acted after another company — later reported to be Amazon, Anthropic’s largest backer — warned that researchers had found a method to extract restricted cyberattack guidance from Mythos 5. Anthropic’s CEO Dario Amodei pushed back, describing the concern as a narrow potential issue that did not justify pulling access to the models broadly, and noting that comparable capabilities were already present in other deployed models. The government did not publicly release the directive or its technical reasoning.
Because Anthropic could not reliably screen users by nationality in real time — a task that would have meant dismissing its own foreign-national staff, the company pulled both models for everyone. Claude Opus 4.8 and other models remained online, but Fable 5 and Mythos 5 went dark globally within hours of the order.
“Overnight, the world’s largest single market — our single market, comprising 450 million people — was cut off from a cutting-edge innovation,” Pröll wrote. “Not by our decision. But by that of a foreign government. This decision was not to be dismissed as a footnote. It was a show of power.”
A partial restoration arrived on June 26, when Washington eased the block for more than 100 approved US institutions. Fable 5 remains fully unavailable. Access for European organisations, including the European Commission itself, which had been in discussions to join Anthropic’s Project Glasswing cybersecurity initiative, had not resumed as of the time of writing.
What Austria Is Actually Proposing
The framing of Pröll’s letter as an invitation for Anthropic to “move to Europe” understates what is really being proposed. The goal is not to relocate a company’s headquarters. It is to relocate the legal jurisdiction that controls access to its most powerful systems.
If Anthropic were established within the EU — even partially, through a European subsidiary with genuine operational independence, US export-control law would have a significantly weaker claim on its models. The kill switch that Washington exercised on June 12 would no longer reach European deployments in the same way. As analysis from FourWeekMBA put it, Austria is not trying to buy a model; it is trying to relocate the permission.
The offer on the table includes three elements: legal certainty, meaning clear EU regulatory frameworks rather than the patchwork of US government discretion; market access, meaning the 450-million-person single market as a commercial incentive; and capital, though Pröll conspicuously did not name a figure or a funding source. The European Commission has been developing a Cloud and AI Development Act as part of a wider technological sovereignty package, and the Austria proposal fits that political direction.
The European open-source AI moment has been building for some time, with member states increasingly aware that their digital infrastructure runs on systems they do not control. The Anthropic episode made that vulnerability visible to politicians and business leaders in a way that academic arguments about digital sovereignty had not.
Why It Is Probably Not Going to Happen — And Why That Is Not the Point
The practical barriers to relocating Anthropic are substantial. The company has announced a fifty-billion-dollar investment in US computing infrastructure, including data centres with Fluidstack in Texas and New York. It has committed more than one hundred billion dollars over the next decade to Amazon Web Services technologies and has secured up to five gigawatts of new compute capacity through that partnership. Amazon has invested thirteen billion dollars in Anthropic and remains its primary cloud infrastructure partner. Anthropic is simultaneously preparing a US stock-market listing and has filed a confidential IPO application.
Europe, for all its regulatory clout, does not have the compute base, the capital depth, or the power infrastructure to replicate what Anthropic has access to in the United States. That gap does not close quickly.
Anthropic itself has not formally responded to the proposal. The company has been expanding its European footprint — hiring a senior executive from the telecoms sector to lead European operations and engaging with EU cybersecurity agency ENISA over the Mythos model — but there is no indication that a structural relocation is under consideration.
Yet treating the Austria letter as simply impractical misses what is happening. The episode follows a wider pattern of US government influence over AI model availability that has alarmed governments outside America. The Five Eyes joint cybersecurity warning issued earlier this month noted that AI capabilities were becoming critical national security infrastructure. Days later, those same capabilities became subject to unilateral US export control decisions that European governments had no role in shaping and no mechanism to challenge.
The pattern extends beyond Anthropic. OpenAI released its new GPT-5.6 model family on June 26 in a government-gated limited preview, with European organisations absent from the initial access list. The architecture is becoming clear: the most capable AI systems are US-developed, US-hosted, and US-controlled, with access to the rest of the world conditional on Washington’s assessment of national security risk at any given moment.
The Broader Context: AI Sovereignty Becomes a Policy Priority
The Anthropic episode lands at a moment when the relationship between the United States and Europe on technology policy is under significant strain. The Trump administration’s tariff policy had already exposed the extent of European economic dependence on American markets. The Fable 5 and Mythos 5 shutdown demonstrated a second, less visible form of dependence: on American AI infrastructure.
The dynamic is structurally similar to what happened with Nvidia chip export controls directed at China — except that in this case, the access restriction affected not an adversary but Washington’s closest allies. The UK, whose government had been in discussions about Mythos access, found itself in the same position as the rest of the world: cut off without consultation.
Pröll’s letter noted that Anthropic had been in talks with ENISA for nearly two months before the June 12 directive arrived and shut those conversations down. European institutions had been treated as legitimate partners until Washington decided otherwise. There was no notice, no transition period, and no alternative path offered.
The EU’s response so far has been measured. A Commission spokesperson said Europe should have a say in how it cooperates on AI security risks. Whether the Commission sees Austria’s proposal as a workable plan or as a political signal should become clearer in the coming weeks as internal discussions about the Cloud and AI Development Act progress.
What This Means for You
If you use Claude in a professional or commercial capacity outside the United States, the events of June 12 demonstrated that your access to the most capable AI tools can be revoked by a foreign government decision you have no visibility into, no vote on, and no recourse against. That is not a bug in the system; it is a feature of how AI model infrastructure is currently structured.
For businesses building workflows or products that depend on frontier AI models, the June episode is a supply-chain risk in the boardroom sense of that term. The Austria letter did not solve it. But it named it precisely for the first time: in 2026, the scarcest resource in artificial intelligence is not compute or talent. It is permission.
Whether the EU can build or attract enough of its own AI infrastructure to give European users a meaningful alternative remains an open question. What is no longer open is whether European governments are paying attention. They are, and they are beginning to act accordingly.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and geopolitics. LiveAIWire publishes daily AI news and analysis at liveaiwire.com.