What Export Controls on Anthropic’s Most Advanced Models Mean for Europe

US restrictions on frontier AI would have come eventually, but few expected sudden export controls. They could be Europe's wake-up call on AI sovereignty.

Afek Shamir — June 19, 2026

On June 12, the Trump administration issued an order requiring Anthropic to suspend access to its two most advanced AI models—Fable 5 and Mythos 5—for non-American nationals, just days after their public release. The decision caused immediate backlash across Europe. Politicians from France’s Gabriel Attal and Jordan Bardella to the Netherlands’ Geert Wilders, alongside European industry and civil society, amplified calls for AI sovereignty. Attal, the presidential candidate for Macron’s Renaissance party, even likened the shutdown of Anthropic’s models to Iran’s blockade of the Strait of Hormuz.

While such reactions are understandable, they underscore what we already know: Europe is behind on AI, overly dependent on the US, and vulnerable to unilateral decisions. Perhaps the renewed sovereignty rhetoric can help build the political capital needed to fix the continent’s AI positioning, but what Europe needs more urgently is a clear-eyed account of what to actually do. In the wake of the Anthropic episode, this piece separates what has genuinely changed from what is being overstated, while identifying where European policymakers should focus their attention.

Europe Should Have Planned for This

In recent years, AI has become increasingly relevant to national security. Europe could have anticipated that the US government would restrict access to US AI models at some point, but has not acted quickly enough to secure its position.

Export controls on US technology have affected Europe before. The use of export controls on American technology for national security reasons is nothing new and is not unique to the current US government. The Biden administration’s AI Diffusion Rule did the same for advanced chips, treating AI hardware as a national security instrument and using export controls to manage its global distribution. Europe was affected then, too, with Tier 1 countries in western Europe split from Tier 2 countries across much of eastern Europe. What the new controls do, less selectively than the Diffusion Rule, is extend export controls from AI hardware to AI models themselves.

The US implemented the new control suddenly and with opaque reasoning. While export controls are not unprecedented, the Trump administration’s recent directive differed from Biden’s Diffusion Rule. This decision happened more quickly, and was driven by harder-to-interpret motives, particularly in light of the recent clash between Anthropic and the Pentagon. It is plausible that the US government wanted to block Fable’s deployment for everyone (Americans included) and used export controls as a tool to do so.

Europe has been moving too slowly to build sovereign AI or secure access to frontier American AI. Europe could have planned more effectively for this moment by treating AI as a sovereign imperative and developing the resources and infrastructure needed to better serve this goal. Instead, the scaling down of EU plans for AI gigafactories and the successive delays to its Cloud and AI Development Act point to the complexity of moving at the pace of developments across the Atlantic. A separate but related issue is the need to access American frontier AI models for, among other things, hardening European infrastructure against cyberattacks. European leaders could have sought to negotiate guaranteed early access to highly capable systems. Yet the EU has lagged here as well, obtaining access to Anthropic’s Mythos model about two months after the company first started sharing it with a small group of American organizations to bolster cybersecurity.

The Restrictions’ Short-Term Impact on Europe May Be Limited

Europe is not, for now, significantly worse off without access to Fable 5 and Mythos 5. This is for two reasons.

Access to Fable 5 and Mythos 5 will probably be restored. The US government and Anthropic will likely come to some sort of agreement to enable the continued rollout of these models. For one thing, the export restriction also applies to foreign nationals working inside American labs—employees crucial to model development. Maintaining the controls would be akin to telling Anthropic to stop developing future AI models.

Other models are good enough for most current applications of AI. Other than some cyberdefense uses, most tasks across the European economy do not currently require the most capable Anthropic models. Alternative models can perform many of the tasks that users would have assigned to Fable, at a similar level of competence. OpenAI’s GPT 5.5, whose capabilities are comparable to those of Fable, remains free of export controls. Open-source alternatives, only months behind, also remain available to European markets. Choosing an AI model is ultimately a practical calculation: the right capability, at the right price for the task at hand. That is why AI adopters often prefer cheaper Chinese models over frontier American ones: they choose based on the job they need to complete, rather than simply selecting the models that sit on top of a capability leaderboard.

Hence, the economic drag Europe will face by losing access to Fable is likely to be modest in the short run. It will compound only if the restriction on European access to American frontier models continues and the capability gap between American models and alternatives significantly widens, neither of which seem likely in the near term.

In the long term, Europe needs frontier AI for cybersecurity and economic competitiveness. Even if the model restrictions’ negative impact can be absorbed for now, the European market will suffer if it cannot access leading models like Mythos and Fable. Aura Salla, a Member of the European Parliament, has argued that “Europe is better off without these models,” because they pose significant cybersecurity risks, with insufficient safeguards. But access to frontier models is an absolute prerequisite for defensive security, allowing domestic companies and governments to identify and patch vulnerabilities before attackers find and exploit them. In the long term, frontier access is also an economic necessity to prevent Europe’s industrial base from being relegated to secondary tiers of productivity. Being locked out of the most capable AI models is categorically different from Europe choosing how to adopt AI on its own terms.

In the Long Term, These Restrictions May Provide Opportunities for Europe

Counterintuitively, the US government’s recent move may turn out to be good for Europe. While a gradual loss of access to American frontier AI might have been anticipated, that eventuality may never have felt urgent enough to prompt serious action. This sudden, unexpected loss of access will be a sobering warning.

The US government’s directive underlined its power to withdraw access. Export-controlling Anthropic’s leading models revealed who holds the kill switch—and how easy it is to press. Even if the controls are overturned, the intent and execution is visible to every government in the world. The use of export controls to restrict global access to AI models signals the distinct power of the American executive branch in shaping the trajectory of AI.

The more dependent Europe is on American AI, the more damaging restrictions could be. Fortunately, for now, Europe enjoys a narrow window of insulation: unlike cloud services, leading American AI models have not yet been deeply woven into European public administration or critical infrastructure. Access to American AI, like access to the nuclear umbrella, may come at a cost to those that accept and depend on it.

The episode is an opportunity to drive political momentum toward middle power coordination on AI. Even if the US directive brings only limited short-term consequences, it illustrates how vulnerable Europe could be in the future if it does not start working to secure frontier AI access now. Countries within the EU, including France, Germany, and the Netherlands, could do more to engage other middle powers like the UK, India, South Korea, Japan, and Canada. They can coordinate on how they wish to govern this technology, secure access to the frontier, and shape how AI affects society. Even if these countries play different roles in the AI stack (and even though existing coordination channels between these countries are currently scant), they share a common interest in preventing any one government from dictating access and governance unilaterally.

As part of a coordinated agenda on AI governance, middle powers could align on their evaluations regimes and incident monitoring practices, while slowly building up shared leverage through collective procurement standards and investment in each other’s ecosystems. A recent example is Aleph Alpha and Cohere’s merger. More ambitious moves could draw on the model of the Eurofighter Typhoon program: pooling procurement across multiple governments to build collective infrastructure or capabilities that no single country could generate alone.

Europe must also navigate public-private partnerships between Washington and American frontier labs. The Mythos episode marks a turning point in government-industry relationships on AI. Despite having been working with government officials and American industry to secure the world’s most critical software, Anthropic had to take its leading models offline immediately at Washington’s request. The age of governments ignoring capable model releases is likely over. The question is whether oversight will be principled and based on scientific thresholds being crossed, or reactive and politically driven. For now, absent federal laws on AI, the latter seems more likely.

The right response everywhere is to invest in building a more mature evaluation ecosystem, enforce actual regulations (like the EU AI Act), and establish regular and transparent predeployment engagement between governments and AI developers—so that policymakers are not scrambling to assess model capabilities in the days after a release. Beyond nurturing positive oversight mechanisms, European governments would also do well to consider how to respond to a reality where consequential AI decisions are made in closed-door conversations between the US government and US AI companies.

How Does Europe Build Leverage When It Needs Results Now?

AI is progressing rapidly, and Europe must act quickly to improve its position. Fortunately, it has a number of tools at its disposal to maintain access to frontier AI models.

Leading AI companies need compute capacity, and Europe could provide it. First, Europe could seek to build compute capacity quickly, along with the energy infrastructure necessary to power it. The aim would be to use this compute not exclusively for European AI development but as infrastructure that reduces dependence and creates negotiating leverage. The constraint on frontier AI development and deployment is increasingly compute, and evidence of its scarcity keeps accumulating. In March 2026, Anthropic tightened peak-hour session limits for paying users due to capacity constraints. By May, it had agreed to spend $1.25 billion per month to rent capacity from xAI’s Colossus cluster through 2029. When a leading AI lab is forced to buy infrastructure from a direct competitor just to keep its basic tiers online, it is a clear signal that compute remains a constraint to AI development.

An AI lab that needs European data centers, European talent, and European revenue has reasons to treat Europe as a partner rather than a market it can afford to work around. If Europe becomes a major host of leading models, this makes restrictions less likely due to the costs to American companies of losing access to the European market. Not so long ago, Nvidia was allowed to sell chips to China because of its importance as a market. To build such leverage in practice, Europe needs to urgently address the barriers to buildout: high energy costs, slow planning processes, and fragmented capital markets.

Middle powers could cooperate more deeply to demonstrate their importance to AI development. Europe should work with other middle powers to mobilize a coordinated response. The semiconductor supply chain runs through the Netherlands, Japan, South Korea, and Taiwan. AI evaluation and testing capacity is concentrated in the UK. Many of the most commercially significant AI applications—from Lovable in Sweden to leading adopters across Canada and India—are built on top of American frontier models, generating the revenue and usage data those labs depend on. Export-controlling American models while depending on allied components, talent, testing infrastructure, revenue, and data should be a strategy with a limited shelf life. Middle powers need to work more closely to demonstrate that they are as necessary to producing AI as the companies currently building it.

Europe must both invest in AI sovereignty and build relationships with US AI companies. Finally, Europe should

amplify efforts to build indigenous frontier capabilities and chips, even if the near-term prospects appear improbable. An ambitious European AI effort would require spending levels that dwarf anything Europe can credibly mobilize through subsidies alone. Meta will invest $125 billion in capital expenditures this year, more than Germany’s entire defense budget, and its AI models still fail to match capabilities from OpenAI and Anthropic. But, difficulty aside, meaningful European AI development will become a necessity over the long term. The question is how Europe can build sovereign AI capabilities while strategically developing partnerships with value-aligned US companies, such as Anthropic. Doing neither or only one has too many downsides; doing both buys optionality.

Export Controls Are a Wake-Up Call for Europe

Arguably, the export controls on Fable 5 and Mythos 5 do not fundamentally change Europe’s position on AI. Europe was dependent on American AI models and chips before last Friday and remains so today. The immediate economic consequences are likely manageable, particularly if the restrictions prove short-lived. But Europe’s response must nonetheless move beyond the familiar debates on sovereignty. It should be concrete: expand compute capacity, coordinate closely with other middle powers, invest in domestic capabilities, and deepen partnerships with firms that need European markets as much as Europe needs them. The overall impact of the recent order could in fact be positive for Europe—but only if it spurs the continent into acting on AI in ways that it already needed to.

---

See things differently? AI Frontiers welcomes expert insights, thoughtful critiques, and fresh perspectives. Send us your pitch.

---

Afek Shamir is an Analyst at RAND Europe, working with the Center on AI, Security, and Technology and the Frontiers of Technology hub. His research focuses on Europe’s role in frontier AI development and governance, with particular interest in the continent’s geopolitical positioning and leverage vis-a-vis other leading AI powers. Prior to RAND, Afek worked at a Brussels-based think tank on the EU AI Act’s governance of general-purpose AI as a Talos fellow and interned at the Tony Blair Institute. Afek holds an M.Sc. in European and International Public Policy from the London School of Economics.