The Trump administration’s sudden decision on 12 June, requiring Anthropic to suspend access to its most advanced models – Fable 5 and Mythos 5 – for people outside the United States, hit the global technology market like the digital equivalent of a strategic strait blockade. Just a few days after its much-hyped launch, the technology that was set to define a new phase in the development of AI became inaccessible to Europeans.

This event marks a critical juncture, a sort of ‘wake-up call’ for European decision-makers. Although the reactions of politicians, from Gabriel Attal to Geert Wilders, can be alarmist, they should be viewed through the dispassionate lens of a trend analyst. This is not merely a diplomatic crisis, but above all a necessary wake-up call for European digital sovereignty.

The Anthropic incident is a painful but valuable lesson. It shows that Europe can no longer rely on the goodwill of its transatlantic partners if it wishes to maintain its autonomy in the age of algorithms. This is a wake-up call which – if properly harnessed – could force a genuine overhaul of the architecture of European technological security.

Lesson 1: The American ‘kill switch’ is a real possibility and closer than we think

The sudden restriction on exports of the Fable 5 and Mythos 5 models has revealed the harsh truth about who really controls the flow of innovation. For years, Europe had assumed that export controls mainly concerned hardware – advanced chips and lithography machines. The Anthropic case proves that the focus has shifted to software itself, and specifically to model weights. These parameters are the ‘brain’ of the AI; the system’s intelligence is encoded within them.

The Trump administration’s decision did not come out of the blue – it was the result of a growing dispute between Anthropic and the Pentagon. This sends a key signal: ‘safety switches’ can be triggered not only for strategic reasons, but also as a result of internal political and business tensions within the US. Control has shifted from physical processors to abstract algorithms, giving Washington a tool to unilaterally shape the global AI landscape.

“The shutdown of Anthropic’s models can be compared to Iran’s blockade of the Strait of Hormuz.” – Gabriel Attal

Lesson 2: The paradoxical absence of pain, or why Europe is safe in the short term

Despite the political uproar, being cut off from Anthropic’s models has not brought the European economy to an immediate standstill. This is down to the market’s pragmatism. At present, most business tasks do not require the absolute peak performance of the latest-generation models.

Europe still has alternatives. OpenAI’s GPT 5.5 model, with capabilities comparable to those of Fable 5, remains available for the time being. Open-source solutions lag behind the market leaders by just a few months, which is an acceptable gap for many sectors. What is more, European companies often make purely economic decisions, choosing cheaper Chinese models for specific tasks, rather than the most expensive American pioneers. Today’s market is a game of good value for money, rather than just about coming top in the benchmark rankings.

Lesson 3: AI is the new nuclear umbrella (with all that entails)

Europe’s dependence on American artificial intelligence is beginning to resemble the concept of nuclear umbrella. It gives a false sense of security and access to state-of-the-art tools, but generates enormous sovereignty costs. By accepting this role, Europe is relinquishing control over key resources for the future.

We are currently in a brief ‘window of opportunity’. Unlike mature cloud services, leading AI models have not yet been deeply integrated into critical infrastructure or national administrative systems. This gives us time to build up our own capabilities before our dependence on the decisions of the US regulator becomes irreversible and begins to paralyse the functioning of European states in crisis situations.

Lesson 4: Infrastructure as a bargaining chip

The most surprising conclusion drawn from the article is that Europe could gain an advantage by becoming a hub for computing power and green energy. It is infrastructure that is currently the bottleneck in AI development. Proof of this is the desperation of Anthropic itself, which, faced with session limits (introduced in March 2026), has decided to pay its competitor – xAI – as much as $1.25 billion a month for the rental of computing power.

If Europe speeds up the construction of data centres and secures a supply of cheap energy, the continent’s negotiating position will change dramatically. It will be much harder for the American giants to ‘bypass’ a market where they physically host their models and from which they draw talent. Having data centres on the ground makes it more difficult to unilaterally shut down the models for logistical and economic reasons.

An AI laboratory that relies on European data centres, European specialists and European revenue has good reason to treat Europe as a partner, rather than merely as a market that can be ignored with a single decree.

Lesson 5: ‘Eurofighter for AI’ – the strength of middle powers

Individual European countries’ solitary struggle for dominance in AI is doomed to failure. The scale of the challenge is best illustrated by the figures: Meta’s planned investments in AI amount to $125 billion a year – this figure exceeds Germany’s entire defence budget. Developing sovereign AI is no longer merely a technological project, but a strategic challenge on a par with the programme Eurofighter Typhoon.

The solution lies in close coordination between ‘middle powers’. Europe (France, Germany, the Netherlands) must strengthen its cooperation with partners such as the UK (a leader in AI safety testing), the Netherlands and Taiwan (key players in lithography and chip manufacturing), as well as Japan and South Korea. Common public procurement standards and investment in ecosystems – as signalled by the merger Aleph Alpha i Cohere – they can prevent any government from unilaterally dictating terms. Only by acting as a united bloc can these countries demonstrate that they are just as essential to the development of AI as the companies that build these models.

Summary and a question for the future

The Anthropic case should not be a cause for panic, but rather a catalyst for constructive action. It demonstrates that Europe must simultaneously develop partnerships with US companies and build its own independent technological base. Investment in computing power, the standardisation of model evaluation systems and the coordination of public procurement are the foundations that will enable us to avoid becoming a passive customer.

We are faced with a fundamental choice that will define our position in the coming decades: Does Europe prefer to pay for access to other countries’ technologies, at the risk of having them suddenly cut off, or will it invest in its own security and sovereignty before the digital ‘kill switch’ is used again?