EU AI sovereigntyfor whom, to what end, and to whose benefit?

The EU AI Act passed into law in 2024 and was celebrated across European capitals as a landmark assertion of digital sovereignty. But Daniel Mügge, writing in the Journal of European Public Policy, asks a question that the celebrations largely skipped: sovereign for whom, exactly? It turns out that this question is the whole story. The phrase "AI sovereignty" entered EU policy vocabulary when Commission President Ursula von der Leyen made "digital sovereignty" a guiding principle in 2020. It has since migrated into AI policy with remarkable speed — and remarkable vagueness. Mügge's core argument is that the vagueness is not accidental. "AI sovereignty" is discursively attractive precisely because it can justify very different, sometimes contradictory, policy goals while appearing to point in one direction. You can invoke it to protect European citizens from Big Tech, or to help European firms compete with American ones. Those are not the same project. In fact, they often pull against each other. Because the phrase papers over that tension, it allows politicians to avoid making a choice. To cut through the fog, Mügge proposes three analytical trade-offs. First: is sovereignty about the EU as a geopolitical actor asserting independence from foreign powers, or is it about citizens asserting control over powerful corporations? Second: is the goal to win a global AI race, or to resist the very logic of that race? Third: does EU AI sovereignty serve Europeans alone, or does it accept responsibility for the rest of the world? Mügge then maps the Commission's AI strategy documents since 2018 and the AI Act negotiations onto all three dimensions. What he finds is consistent across all three, and it isn't flattering. Starting with the first trade-off, jurisdictional sovereignty, as Mügge defines it, means "the relative independence of a geographically circumscribed unit from other major powers." In practice, it shows up as calls for "AI made in Europe," investment in European AI infrastructure, and regulatory independence so that European prosperity doesn't hinge on decisions made in Silicon Valley or Beijing. The logic has real grounding; AI's military and economic relevance gives states genuine incentives to protect their autonomy. Citizen sovereignty is a different concept. It treats sovereignty as "a form of legitimate, controlling authority" held by the public and locates the main threat not in foreign states but in large domestic and foreign corporations. From this perspective, the problem is private extraction, algorithmic harm, and technologies that disproportionately benefit capital over labor. Citizen sovereignty would require tightly constraining powerful private actors, not promoting them. Mügge's document analysis finds the EU systematically picking the first option. Commission strategy texts up to 2021 covered a broad governance agenda, but as the AI Act proposal crystallized after April 2021, the discourse narrowed toward regulatory intervention, and initiatives like the Strategic Initiative on AI Startups and Innovation quietly faded. The final AI Act resembles the Commission's original proposal closely. The European Parliament did add language emphasizing environmental and societal impacts and uneven distribution of AI's benefits, but it also proposed language stressing "AI made in Europe." The Council, meanwhile, pushed to ensure regulation wouldn't become "disproportionately burdensome for companies." The jurisdictional framing prevailed. The idea of corporate power as a problem worth directly confronting did not survive intact. The second trade-off — racing versus resisting — tells a similar story. Mügge's reading of Commission communications reveals a consistent embrace of the race framing. As early as 2018, the Commission warned that "European industry cannot miss the train" and argued it was time to ensure "Europe is competitive in the AI landscape." The regulatory framework was designed, in the Commission's own words, "to intervene only where this is strictly needed and in a way that minimizes the burden for economic operators." Calls for "the broadest possible uptake of AI in the economy" and for "unlocking" data run through document after document. Distributional concerns receive rhetorical acknowledgment and little else. The strategy pledges that "no one is left behind in the digital transformation," yet the concrete proposals are retraining schemes to align workers with business needs — not measures to slow diffusion or compensate those displaced. The literature Mügge cites estimates that ten percent or more of jobs could be lost within a decade. The Commission's answer is skills policy, not a rethink of the pace. The AI Act negotiations over foundation models brought the competitiveness logic into sharp relief. The European Parliament remained cautious about the most powerful AI systems. But Commission Vice-President Vera Jourová warned in October 2023 against being "paranoid" about regulating generative AI. Around the same time, France, Germany, and Italy circulated a non-paper warning that over-regulation would hinder "Europe's ability to innovate," motivated by the rise of European AI startups like Mistral and Aleph Alpha, each having raised several hundred million euros. Those competitive pressures nearly derailed the AI Act entirely until a compromise was reached just days before the final vote. This episode shows how thoroughly the race logic had taken over the deliberations. An emancipatory alternative would have looked different. Mügge describes it as using AI sovereignty to deliberately slow diffusion to protect jobs or to sacrifice some competitive advantage to pursue a genuinely distinct European path. That path was theoretically available. The evidence shows it was not taken. The third trade-off is where Mügge delivers his harshest verdict. EU AI policy not only prioritizes European interests over citizen interests but also largely ignores people outside the EU altogether. References to countries beyond the main AI competitors are sparse in Commission documents. The dominant assumption is that if Europe builds ethically superior AI, "everyone will benefit from the ethical superiority of 'AI made in Europe.'" The benefits are presumed to radiate outward automatically. No mechanism is examined. No concrete harms to poorer countries are weighed. Mügge uses agriculture as a concrete illustration of this logic. Commission texts promise that AI-powered precision farming will "reduce pesticides, fertiliser and water consumption" and that AI "is destined to make impactful contributions to global challenges as well as development policy." The framing positions poorer countries as passive beneficiaries — customers or recipients — rather than actors whose perspectives deserve weight in their own right. Meanwhile, the actual policy goal is to make the EU "an assertive player in fair and rule-based international trade," leveraging competitive advantage in AI markets. Global responsibility is just a slogan. Trade advantage is the strategy. The environmental dimension, which is inherently global, gets acknowledged mainly through concerns about data center energy use. The response is to pursue low-energy computing research — again to retain competitive edge — not to question the pace of AI development itself. As geopolitical tensions have risen, the earlier idealism about Europe's ethical leadership has quietly faded. After the US executive order on AI policy in October 2023 echoed many issues that EU negotiators had framed as distinctively European, the claim that European ethics alone would yield global good became harder to maintain. What ties all three findings together is the paper's closing argument, and it is the one that matters most. EU AI sovereignty is not a natural response to technological change. It is a set of deliberate political choices. Mügge insists on this point because there is a pervasive tendency in tech policy debates to treat the shape of AI development as given — something that arrived from outside politics and that policy must adapt to. His analysis of six years of Commission strategy documents and legislative negotiations shows otherwise. The decisions about who counts as sovereign, whether to race or resist, and whose interests get considered are choices. They were made by identifiable actors under identifiable pressures. Different choices were available. That matters because framing determines who sits at the table and which trade-offs become visible. A sovereignty framing centered on jurisdictional independence keeps the conversation focused on Europe versus the world, and keeps European firms as allies rather than subjects of regulation. A citizen sovereignty framing would invert both. The EU chose the first. The AI Act that emerged reflects that choice — in what it regulates, what it soft-pedals, and whose interests it was designed to serve. The precondition for achieving different outcomes is simply recognizing that this is politics, not inevitability. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.