When access to frontier AI becomes a geopolitical switch

In early June, within days of Anthropic launching Fable 5 and Mythos 5, the US government imposed export controls on the models, citing national security concerns. The decision prompted a wave of concern about Europe’s dependence on U.S. tech companies. Reporting suggests that the foreign-national restriction was not simply a targeted move against Europe, but a fast and legally available way to force a broader pause in access. Anthropic says the order required it to disable the models for all customers to ensure compliance. Whether this decision is justified or reversed is not the main point. The point is that access to frontier AI can become a question of geopolitical discretion overnight.

This is precisely the kind of moment the newly launched Europe 2031 scenario was written for, a Europe that depends on systems it does not control, and then discovers the depth of that dependency only when access is removed. The scenario is not meant as a prediction, but as a warning about a plausible future in which Europe all but loses the ability to chart its own path. 

Why Europe must protect its ability to choose

The Anthropic case points to a broader shift in the nature of technological choice. With decentralised technologies, such as writing, printing or email, no single gatekeeper can easily withdraw the choice to use them. Frontier AI inverts that logic. Access often runs through centralised chokepoints, from infrastructure and supply chains to platforms, compute and export controls. The question is therefore not only whether Europe chooses to use a technology, but whether the choice to use it remains available at all.

If technological choice can be withdrawn, technological sovereignty must be understood as the ability to keep choice open. Europe must be able to decide which technologies it uses, under what conditions, and for which public purposes. This choice will not be secured by declaring sovereignty, nor by copying the technological trajectories of others. It will be secured by building the capacities that keep future choices open.

No technology is an isolated tool. Each rests on energy, capital, supply chains, skilled people and institutions. Our choice is therefore never a binary decision between accepting or rejecting technology, it is always shaped by earlier choices.

This is why the distinction between present and future sovereignty matters. Present sovereignty secures what already exists; future sovereignty builds what does not yet exist. Present sovereignty is often a report card on past choices. Future sovereignty is the effort to make the next report card look different.

Making interdependence negotiable

Europe does not need to reproduce the path of the United States or China. It needs the capacities to choose where to depend, where to compete or collaborate, and where it can lead. In practice that implies affordable energy, compute, mobilised capital, global talent, and the organisational ability to deploy AI where Europe already has depth. It also means treating data centres as critical infrastructure and building institutions that learn through use. And it entails a positive vision of what AI and other emerging technologies can do for society.

Full technological independence is impossible, and it is not desirable. The United States depends on global talent, Taiwanese chips, Dutch lithography, European enterprise software and complex supply chains. China and the US remain mutually dependent in ways neither side can fully unwind. Europe should not seek autarky, but enough technological leverage to make interdependence reciprocal rather than one-sided. 

The same logic applies to procurement. Buying European for its own sake is not enough. If “European” means less capable or less secure systems, it weakens sovereignty. The aim is to procure for capability, resilience and bargaining power, not for origin alone. Procurement can also protect that choice. Contracts for critical technologies could include continued access, portability requirements, fallback access to essential assets if a service is withdrawn, and real penalties when procured services are made unavailable. Such terms would not end dependency, but they would make it more negotiated and enforceable.

The U.S. restrictions on Fable 5 and Mythos 5 should not be mistaken for the end of Europe’s technological choices, but they are a warning that the ability to shape technology depends on capacities built over time. Anthropic will not be the last company, nor AI the last technology, to make Europe confront whether its choices remain available at all.

Europe’s long game for technological freedom

Europe has to play the longer game, building application power in sectors where it already has depth, and investing early where strategic value is high and technological paths remain open. Being a significant user creates leverage, but mainly with commercial providers that want to keep their customers. It does less when a government acts on security grounds, as Anthropic showed.

That is why continuity terms matter, alongside the ability to run different models for different tasks. The decisive question is not only who builds the most capable models or data centres. It is who can choose between systems, use them at scale, learn from that use, and keep alternatives ready. That is what keeps future choices open.

Read more about the authors.

Nicklas Berild Lundblad is a writer, researcher and public policy expert.

Johanna Ballesteros is a Research Associate at the Oxford Internet Institute.