European AI Policy Needs Cognitive Sovereignty

European AI policy keeps talking about digital sovereignty like the big question is where the server rack sleeps at night. Meanwhile everyone in the room is drafting with ChatGPT, searching with Google, coding with Copilot, and quietly importing a worldview from somewhere between San Francisco and Seattle.

That’s not sovereignty. That’s outsourcing with a flag on top.

I felt this hard reading the recent Wired Italia argument about “sovranità digitale senza sovranità cognitiva.” The point is almost offensively simple: if Europe controls some infrastructure but not the systems shaping knowledge, interpretation, and judgment, then European digital sovereignty is half-built at best and pure theater at worst.

And before someone accuses me of doing anti-American drama from a café in Milan, relax. I live in America. I build in tech. I use American AI products constantly. That’s exactly why this bothers me. I know how fast a useful tool turns into a dependency you stop noticing.

First it’s a tool.

Then it’s your workflow.

Then, if you’re not careful, it starts organizing how you think.

European AI Policy Is Protecting the Pipes and Ignoring the Brain

A lot of European AI policy still treats sovereignty like a hardware problem. Cloud. Chips. Data centers. Cybersecurity. Procurement. All important. All necessary. Also all very photogenic for politicians standing in front of a blue backdrop with twelve stars and a logo the size of a Fiat.

But the real fight is one layer up.

If the systems Europeans use to read, summarize, rank, search, write, and decide are foreign by default, then the part that shapes judgment is foreign too. You can host the server in Frankfurt and still have the epistemology shipped from California.

That’s the uncomfortable bit. It’s easier to cut a ribbon at a semiconductor facility than to admit AI systems are becoming default interpreters of reality. They don’t just fetch information. They compress it. Frame it. Prioritize it. They tell people not only what exists, but what matters.

That’s why the Wired Italia thesis lands. Hosting data is not the same thing as shaping meaning.

My nonna would put it more clearly: if someone else picks the ingredients and cooks the sauce, don’t call it your ragù.

Europe has made real progress on infrastructure. Gaia-X matters. EuroHPC matters. Semiconductor policy matters. Cyber rules matter. I’m not dismissing any of that. But if the interfaces through which Europeans think are imported, then EU AI sovereignty is still mostly a lower-layer story.

And habits form at the top, not the bottom.

Convenience Is How Dependency Sneaks In

Nobody in a startup meeting asks, “What is the most philosophically sovereign stack for our civilization?” They ask, “What works by Friday?” If cash is tight, they ask what works by tonight.

That’s how dependence happens. Not through conquest. Through convenience.

One API here. One copiloting tool there. One enterprise contract later. Suddenly an entire team can’t function without non-European models. Then the switching cost stops being technical. It becomes cultural. People start writing in the rhythm of the tool. Researching in the logic of the tool. Trusting the defaults of the tool because the defaults are fast and weirdly soothing.

We’ve already watched this happen at absurd speed. UBS reported in early 2023 that ChatGPT hit 100 million monthly active users in about two months after launch. Two months. That’s not adoption. That’s a behavioral land grab.

And the model layer is still dominated by US companies. OpenAI. Google. Microsoft. Anthropic. Meta. In most enterprise conversations I hear, those are the default names before anyone even pretends to scan the European field. That’s not because Europe lacks talent. It’s because markets reward immediacy, while sovereignty requires patience, coordination, and a political spine. Three things Europe possesses, theoretically, usually in PowerPoint form.

I say this as a guilty person. Last month in Lisbon I used three AI tools in one afternoon to clean up a deck, summarize a contract, and fix code I absolutely should have written properly the first time. I love a shortcut. I’m Italian. We invented elegant cheating around bad systems. But I could feel the habit forming in real time, and that’s the part that gets me.

Most smart people are not consciously surrendering anything. They’re just tired. Busy. Ambitious. Easily seduced by products that save twenty minutes and remove friction. That’s how power consolidates in tech. Not with a bang. With a better UX.

So when people reduce AI in Europe to “can we host the data locally?” I want to throw my phone into the Arno.

Gently. But with intent.

The Model Is Never Neutral

AI models are not neutral lookup engines. They reflect training data, reinforcement choices, safety tuning, ranking logic, language priorities, and a thousand product decisions nobody outside the company really sees. Even when the output sounds blandly helpful, the system underneath is making calls about relevance, acceptable speech, ambiguity, and truth.

That matters a lot in Europe because Europe is gloriously messy. We are not one monoculture with one legal code and one dominant language. We are 27 member states, 24 official EU languages, and an entire continent of historical baggage, legal nuance, regional identity, and administrative weirdness. I say that lovingly. It’s our charm. It’s also a nightmare for lazy product design.

Ursula von der Leyen said in her State of the Union speech on 13 September 2023 that Europe has become a global leader in AI regulation. She’s right. The EU has absolutely set the pace on rules for trustworthy AI. Good. The world needed at least one adult in the room.

But regulation leadership is not model leadership.

That’s the gap people keep trying not to stare at. Europe talks a lot about values, human-centric technology, trust, safeguards. Fine. I agree with all of that. Deeply. But values do not magically install themselves inside systems Europe does not build, train, or meaningfully control.

That’s my issue with so much AI regulation in Europe discourse. It assumes Europe can be the moral operating system while outsourcing the actual operating systems. Cute idea. Not a durable one.

Mario Draghi’s 2024 report on European competitiveness made the broader point in harsher terms: Europe has to close its innovation gap if it wants to stay economically and politically relevant. AI sits right in the middle of that warning. If the continent becomes a place that mainly regulates technologies invented and scaled elsewhere, then Europe risks turning into a museum curator of values instead of a producer of power.

And please miss me with the “models are neutral because math” line. That’s Silicon Valley bedtime storytelling for adults. Anyone serious about multilingual systems, legal tech, public-sector AI, or education knows the same thing: design choices carry ideology, incentives, and culture with them.

If European media, schools, public administrations, and companies increasingly rely on non-European models, then European norms become downstream of someone else’s roadmap.

That should scare people more than it currently does.

I’m Pro-European. That’s Why I’m Annoyed.

My position is not subtle: the answer is not some sad little techno-nationalism where every member state pretends it can build sovereign AI alone.

Italy alone won’t do it. France alone won’t do it. Germany alone won’t do it. And I say that as an Italian who would love to believe espresso, taste, and chaotic improvisation can solve structural problems. Bellissimo fantasy. Not reality.

The only serious answer is European scale.

Compute at European scale. Procurement at European scale. Shared datasets, multilingual public infrastructure, research funding, startup financing, and industrial policy at European scale. If we keep acting like 27 semi-coherent fiefdoms, we’ll get 27 AI strategies, 27 launch events, 27 glossy PDFs, and approximately zero EU AI sovereignty.

This is why I’m pro-EU on tech. Not because Brussels is flawless — lol, obviously not — but because a united Europe is the only level where this is strategically plausible. The single market exists for a reason. We should probably use it.

Enrico Letta’s 2024 report on the future of the single market made basically that argument in broader economic terms: Europe has to think bigger, integrate more deeply, and stop letting fragmentation kill scale. AI is one of the clearest examples. Maybe the clearest.

And yes, I’m impatient because I’ve seen the American version of scale from the inside. Not always coordinated by government, but coordinated by capital, procurement, platform effects, and a level of ambition that doesn’t apologize for itself every five minutes. Europe has the brains. What it lacks is synchronized conviction.

We have world-class researchers. We have industrial depth. We have institutions that people still expect things from. We have legal traditions worth defending. We have 450 million people in the EU single market. This is not some tiny helpless peninsula with nice museums and seasonal despair.

But we act smaller than we are.

What Cognitive Sovereignty Would Actually Look Like

I don’t mean a philosophy seminar where everyone says “epistemic autonomy” and then disappears for natural wine. I mean practical stuff. Boring stuff. The kind of stuff that actually changes power.

A real cognitive sovereignty agenda would start with European foundation models and open models that are genuinely strong in major EU languages, not just English with subtitles and a polite apology in Slovenian. Legal nuance, administrative language, regional context, public-service use cases — those should be core design inputs, not localization chores.

It would also mean procurement that creates actual demand for European AI models. Not symbolic support. Not another conference with lanyards and croissants. Real contracts. If public administrations say they want sovereignty and then keep buying only from the usual non-European giants, they are not building sovereignty. They are financing dependency with taxpayer money.

Schools matter too, maybe more than people want to admit. If students grow up treating AI outputs as neutral and authoritative summaries of reality, we’re cooked. Cognitive sovereignty means teaching people how models fail, how they skew, how they flatten uncertainty, how to interrogate an answer instead of kneeling before it because it arrived in a confident paragraph.

Same for media. Same for government. Same for companies.

There are reasons for optimism. Mistral is the obvious example — yes, French, but more importantly proof that Europe can produce globally relevant AI companies. There’s also real open-source momentum across the continent, plus serious compute efforts through EuroHPC.

According to the EuroHPC Joint Undertaking and related EU announcements in 2024, Europe has been rolling out AI Factories connected to its supercomputing infrastructure to support startups, researchers, and industry. Good. More of that. Faster. Bigger. Less ceremonial.

Because this is the key thing: cognitive sovereignty is not autarky. I’m not arguing Europe should go offline and pretend America doesn’t exist. That would be stupid, self-harming, and honestly very un-Italian. We take good ideas, remix them, add olive oil, and improve them. That’s half of civilization.

What I want is bargaining power. Alternatives. The ability for Europe to shape its own epistemic environment instead of renting one.

Think of it less like building a wall and more like finally owning a kitchen where we can cook our own food.

Yes, I made it about food. Of course I did.

I don’t want Europe isolated.

I want Europe unembarrassed.

The Question Europe Can’t Dodge

If the next generation of Europeans learns, writes, codes, researches, searches, and argues through systems built elsewhere, what exactly are we calling sovereignty?

That’s not rhetorical. I think it’s the central question for European AI policy over the next five years. The blocs that matter most won’t just be the ones with chips, data centers, or regulation. They’ll be the ones whose AI systems become the default layer through which reality gets organized.

That is a power position.

Maybe the power position.

Europe can still claim it. But not if digital sovereignty in Europe stays a branding exercise about infrastructure while the cognitive layer gets outsourced by habit. Not if we confuse compliance with capability. Not if we keep congratulating ourselves on values while underinvesting in the machinery that makes those values real.

So yes, build the cloud. Fund the compute. Pass the rules. Do all of it.

But if we stop there, we’re not sovereign. We’re just renting independence with better paperwork.

And for a continent with this much history, talent, ego, and unrealized force, that’s honestly a little pathetic.