Mythos and the Coming End of European Digital Sovereignty

Over the past years, Europe has voiced the ambition to become one of the most influential actors globally in governing artificial intelligence (AI). Through the AI Act and the Cyber Resilience Act, under the rubric of digital rulemaking, Brussels engages in standard-setting that must be followed by firms and individuals.

Now, governing a technology is by no means the same as controlling the capabilities that make it powerful, and it is that gap between rule-making authority and material capacity that will decide Europe’s place in the order to come. The European Commission itself now concedes as much.

The Action Plan on Cybersecurity and Artificial Intelligence, published on July 7, acknowledges that frontier capabilities are chiefly developed outside the European Union and that their availability is often determined by non-transparent processes, effectively positioning advanced AI as a matter of Europe’s technological sovereignty.

Europe, in other words, has understood the necessity of building and not just regulating, yet the question remains whether it can build at the speed and scale required. The recent Mythos episode is the clearest illustration of what is at stake and where the issues lie.

In the wake of a sudden and sweeping unilateral restriction on frontier artificial AI based on national security grounds by the US government, high-level meetings were scheduled at the European Commission and the G7 to discuss the strategic fallout of the Mythos incident.

The European Commission has since urgently evaluated the practical implications of the surprising US export control directive, while Anthropic CEO Dario Amodei has attended a G7 working lunch with world leaders to discuss the latest episode of the geopolitical securitization of AI. AI and access to certain high-capability large language models (LLMs) have become a gated strategic resource that is fenced off from those humans who reside in the wrong countries or possess the wrong passports.

When the US government issued an order suspending access to Anthropic’s Fable 5 – an LLM with guardrails that is based on the Mythos model – to foreign nationals on June 12, the New York Times notes that “the order limiting who can use Anthropic’s technology is unusually expansive,” and that “foreign nationals who work for Anthropic are also affected by the suspension,” which includes the firm’s recent star hire: the former OpenAI co-founder Andrej Karpathy. This restriction offers an insight into a nascent form of frontier AI governance, which directly affects Europe’s sovereign enclosure.

The recent US government intervention involving Anthropic’s advanced AI models illustrates this emerging reality. A decision by Washington affecting access to frontier AI systems demonstrates that access to highly capable models may increasingly depend on geopolitical considerations, export controls, and national security decisions. For European policymakers, the episode raises a difficult question: What happens when the technologies shaping economic competitiveness, scientific discovery, and national security are controlled elsewhere? The answer may determine whether Europe becomes a genuine technological power in the age of AI or whether it remains primarily a rule-maker in an ecosystem shaped by others.

Limits of AI governance

This emerging reality exposes a somewhat painful truth about the geopolitical posture of European states. Europe is in the curious situation of being structurally arranged to notice the future, regulate the future, moralize the future, and not be involved in building the future.

In Brussels, the EU institutions regurgitate European sovereignty, as policymakers actively construct safety standards and regulation frameworks to protect EU citizens. But in the AI domain and the endeavor to harness the technology that defines the future of humanity, sovereignty gives precedence to the ability to produce semiconductors, exponentially growing amounts of required energy, data centers, as well as to organize and attract global talent, capital, and deployment channels. Europe and its decision-makers tragically confuse rule-making capacity with future-shaping capacity in AI.

In fairness, Brussels now acknowledges this truth. The EU’s Action Plan on Cybersecurity and Artificial Intelligence is unusually blunt in stating that Europe must “develop its own sovereign general-purpose frontier AI capabilities to mitigate the risk of new dependencies,” and warns that “without compute, models, and data infrastructure, Europe is bound to remain a vulnerable user of frontier AI systems made elsewhere that others can suddenly switch off.”

The Commission appears to have made a 360-degree change on its previous preference to regulate while others build; it is now explicitly, too, talking about capability and infrastructure. In short, the insight is no longer contested even in Brussels, though the execution is unresolved, as it becomes questionable whether a politically fragmented EU can convert this recognition into executable steps.

A critical scenario, titled “Europe 2031,” published by a group of AI researchers just a day prior, on June 11, cautions, “unless we embark on it now, Europe will lose the ability to shape its own future,” though policymakers might utter feelings that are closer to “No more American AI; no American clouds.”

As if that is a win, and has merit all by itself. What is critical to realize in such a scenario — a partial de-risking or full decoupling from US technology — is that this represents a strategically hollow autonomy, which is fundamentally unsupported and unsustainable by material capacity and capability. EU policymakers seem to have forgotten that digital sovereignty cannot simply be legislated into existence through regulations and procurement rules. Digital sovereignty is developed over time through attracting talent that is well paid, a wide regulatory space to experiment, cheap energy, and abundant physical infrastructure like data centers.

The material bottleneck

To understand Europe’s precarious position on AI, one must have a hard look in physical preconditions. The development of AI at rapid speeds requires continuous electricity, vast space or land, industrial cooling, cutting-edge semiconductors, optimized supply chains, and specialists who demand astronomical (for European conditions) salaries. As an industry observer rightly notes, “AI leadership is very much about massive financial capital, massive access to energy and GPUs, and the best world-class talent.”

While Europe has a pipeline of well-trained specialists through numerous universities and the necessary entrepreneurial spirit, too, the foundational conditions required for frontier AI are either overtly hostile or chronically undersupplied and further constrained by bureaucratic inertia. Starting with cheap and reliable power, energy in today’s Europe is mainly seen as an environmental issue that must be solved through costly and non-continuous renewables. The European Commission speaks of the need to achieve “sustainable digitalization,” while external pressures like Russia’s invasion of Ukraine and the blockage of the Strait of Hormuz further complicate energy security in Europe.

Yet, continuous and cheap electricity is the absolute baseline of AI development. And it is required not in ten, twenty, or fifty years, but immediately. Computational capacity, critical to AI development, must scale and thus demands immediate grid access, rapid permitting processes, immense capital concentration, and speedy data-center buildouts. But anybody living in Europe today, for instance in Germany, understands well how a scaled and sustained demand for more (really, much cheaper) energy would send political factions like the Green Party spiraling, and thus, that issue alone would run into domestic party conflict and never-ending debates without resolution.

Global talent inevitably follows the highest salaries and most ambitious AI labs in order to work on meaningful and life-changing projects. But Europe’s labs remain overburdened by structural costs and regulatory constraints, as well as the formidable competition from US-omniscalers and Chinese state-industrial mobilization. In essence, a structurally handicapped Europe is regulating AI faster than it can act on energy security or energy costs to facilitate and support an AI boom.

The coming vassal position of Europe

These material weaknesses inherent to contemporary Europe inexorably lead to geopolitical consequences. If the US ultimately controls the best AI systems, and China commands the advanced industrial frontier, Europe will not emerge as a balancing third pole in the race to the technology of the future. What appears more likely is that European states and their populations will end up as what Axel Voss, a member of the European Parliament, called a “digital colony” of other powers. As a consequence, Europe will continue to lose talent and be reduced to a consumer, regulator, and passive bargaining object. As the scenario predicts for April 2029 (one must mention, only three years from now), “Europe depends on American AI, but the US has no such dependency on Europe.”

Economic dependence then becomes a brutal political reality because someone else possesses the power to ration access to frontier technology, right as the US becomes more transactional by the day, indicating a dangerous mixture.

For March 2031, the scenario grimly notes, “American labs hold the cognitive frontier; China still holds the physical one. Atlas [a fictitious US firm] alone is worth more than every listed European company combined.” One has to acknowledge that in a future where European industry, defense, and scientific research are fundamentally reliant on foreign AI models, the capacity and room for independent actions and decisions will be reduced or vanish.

Europe could well face delayed or no access, country-tiered inference licensing, or the constant threat of having its most critical digital technology throttled by a third party. In such a future, any attempt of European actors to reassert geopolitical independence will be constrained and punished, since the valve to transformative AI (or other technology) is opened or closed in either the US or China. To be blunt, even if Europe aims to strike back in a tit-for-tat, it will lack reciprocal leverage, because in that asymmetrical reality, “obvious retaliation is self-defeating,” which leaves the continent functionally subservient and weak, unable to dictate its own technological future.

Tellingly, the EU Action Plan already anticipates this exact predicament. It instructs that a forthcoming “European Blueprint” for structured access to advanced AI shall include contingency measures for the potential situation where access to a model “is restricted or withdrawn by a provider or a third-country authority.” This, of course, is an implicit admission that a Mythos-type shut-off is not a remote hypothetical but a contingency Brussels is now formally planning around. Ironically, the fact that the Commission is drafting fallback procedures for the day the AI-valve is closed is strong evidence that the vassal position described in this article is already being priced in.

AI as a technological multiplier

Skeptics may well argue that the risks illustrated in the scenario Europe 2031 are overblown and inflated, or viewed through the distorted lens of a US obsession with an imagined race to artificial general or super intelligence (AGI/ASI). However, one does need to take the belief in an imminently arriving omnipotent technology, such as AI, to recognize the universal gravity of Europe’s strategic positioning.

The argument is broader and holds even if AI never achieves the most extreme forecasts of sentience or capabilities beyond what humans can do. As many recognize, AI already proves to radically increase human productivity and advance scientific discovery across most disciplines, such as fundamental science, cyber defense, and advanced robotics, proving its value as a technological multiplier in this early phase of AI development.

Europe’s dependence on the US and China can easily become critical even without AI becoming omnipotent. Crucially, missing the foundations of this technological era means Europe operating at a persistent structural disadvantage in every subsequent arena of global competition, risking permanently bogging down Europe in scientific innovation and making it perpetually vulnerable in defense.

The most tragic aspect of the current trajectory is that the result will not stem from ignorance; the impending crisis is widely understood. But European politics appears to be simply unable to metabolize the writing on the wall. The required corrections to set Europe up for success – continuous and cheap electricity, rapid grid connections, permissive data-center zoning, massive capital concentration – are too monumental and structural for a politically disunited Europe to undertake in time.

All that becomes clearer as wartime-speed industrial modernization due to Russia’s full-scale invasion of Ukraine, even under European unity, has proven insufficient. Vetoing and regulating is the new mantra of European modernization. If nothing fundamental changes in the coming years, historians will note the narrative failure, as predicted for June 2034 in the scenario: “We had a negative vision. We were good at negative visions. (…) The political environment is never ready.” The ultimate indictment might be the realization that “there was a chance and we did not even try.”

Europe’s future is not likely to be decided by military subjugation or other external events; it will much more likely be strongly influenced by AI-linked conditions imposed by third parties, such as conditional licensing, rationing, API-gating, and high prices, with little to no pathways of resistance. Europe’s future might be placed behind a gate that itself no longer controls.

What European success in AI would look like

If diagnosis were the measure of success, Europe would already have won; as a German proverb states, “we do not have a knowledge problem, only an implementation problem.” Europe and its institutions know what the issue is, and success does not require Europe to out-build the US or out-manufacture China. It requires something simple to state and hard to obtain: enough sovereign capability so that access to frontier AI cannot be switched off from the outside without a European fallback ready to soften the blow.

Concretely, this means treating the obviously ambitious commitments in the EU Action Plan as obligations. The EU Commission concedes that building Europe’s frontier capacity “will entail hundreds of billions of euro investment needs which can only be partly covered by public finances” and that although “the cost for Europe of building its own frontier AI capacity may be very large, the cost of not building it may be even larger and grow every year as the AI-capability gap widens.” Thus, private capital would be necessary to close funding gaps.

Nevertheless, execution on the unglamorous foundations described above, such as rapid grid connections, quick permits, fewer regulations, and most critically, cheap energy, remains undisputed. Encouraging, the mentioned “European Blueprint” of structured access is certainly worth pursuing, but it can only be a supplement to sovereign capacity. A fallback plan for having ones AI-valve turned off by outside parties cannot be equated to being un-switch-off-able, which really should be Europe’s ultimate long-term objective. A minimum requirement for European success in AI would be resilience due to the ability to weather a Mythos-styled restriction without European industry, defense, and research grinding to a halt.

Whether Europe and its institutions in Brussels are able to create suitable conditions for success in AI, however, is not decided by the quality of analysis, for instance, in the EU Action Plan, but by political will to encourage and sustain the building of indigenous AI in Europe.