Europe Disconnects Itself from AI

WWDC26, Cupertino. The keynote has just ended, and for two hours we all had the same feeling: Siri AI, the real one, the agentic assistant we had been waiting for since Apple Intelligence was announced in 2024, is finally here. A Siri that can see your screen, write for you, act across your apps, and pick up the thread of a conversation. The moment when the iPhone becomes a promise again instead of just a habit.

And then the press release landed. Three paragraphs later, a footnote. The list of the twenty-seven countries that will not get it on iPhone and iPad. France, alphabetically, stuck between Finland and Germany. I read it inside Apple Park itself, and I had the same thought as many French-speaking developers that evening: we just got unplugged again. Except this time, Apple did not cut the power. We did.

Let me say it straight away, because I know it will make some people grind their teeth: on this one, Apple is right, and Europe is wrong. Not a little wrong. Wrong enough to relegate us, European citizens and developers, to being spectators of our own future.

Your Mac will be smart, your iPhone will not

Let us go back to the facts, because they are almost comical in how absurd they are. Siri AI will indeed come to the European Union — on macOS 27, on visionOS 27, on watchOS 27. Your Mac will have the intelligent Siri. Your watch will have the intelligent Siri. Your €3,500 headset will have the intelligent Siri. But the device you take out fifty times a day, the one that is literally the most used platform on the planet, the iPhone — that one will not. And neither will the iPad.

This is not a bug. It is the direct consequence of the Digital Markets Act. And it is not even the first time: remember, Apple Intelligence was announced in June 2024, rolled out in the United States in October, and only arrived on European iPhones in April 2025 with iOS 18.4. Once again, the Mac was served before the phone. We are watching the exact same movie again, only worse, because this time there is no date. No "be patient, it is coming in April." Craig Federighi said it plainly: Apple currently has no timeline for bringing Siri AI to iOS and iPadOS in the EU.

The reason? The Commission's interpretation of Article 6(7) of the DMA. For Brussels, the moment Apple exposes an assistant feature like Siri AI, interoperability requires opening the same level of access to any third-party assistant. But Siri AI is not a weather widget. It is an agent that reads and sends your messages, accesses your files, makes purchases, and controls your applications. Giving those keys, without safeguards, to any actor that asks for them is exactly what the DMA would require under the Commission's reading. Apple proposed a way out — an intermediary called the Trusted System Agent, plus a gradual eighteen-month rollout to bring everyone up to the necessary security level. The Commission said no. To everything. No counterproposal, no compromise. No.

The real scandal is that we cannot even test it anymore

People talk a lot about users. But there is one sentence in that statement that drove me up the wall, and it should alarm all of us: developers located in the EU will not be able to test or use Siri AI features for their apps on iOS and iPadOS.

Do you understand what that means? I have been coding on Apple platforms since 2007. I shipped my first app shortly after the App Store opened. I built ⌘+F, a community of French-speaking Apple developers, precisely because I believe that here, in France, we have people capable of building incredible things on these platforms. And now I am being told that this whole community — brilliant indies, CTOs, companies that train teams inside some of France's largest groups — will not even get access to the SDK for the most structurally important feature of the decade. You cannot ship what you are not allowed to touch.

The future of the app is the agentic app: applications that Siri knows how to operate, that plug into the assistant's reasoning, that become tools in the hands of AI rather than islands you open one by one. While a developer in San Francisco, London, or Tokyo spends the summer learning, prototyping, breaking things, and preparing for launch day, the developer in Lille, Paris, or Niort will have to... watch. The only way to get your hands on Siri AI from Europe today is to pretend to be American, with a US App Store account — proof, if any were needed, that the wall is not technical: the feature runs perfectly, you just have to lie about where you live. Otherwise, you have to actually move. Talent does not disappear. It relocates. And that is the most expensive kind of drain, because it stays invisible until the day it becomes irreversible.

We have seen this movie before, and it always ends the same way

The wildest part is that none of this is new. The DMA has been fully applicable since March 7, 2024. Since then, the list of features missing or delayed in Europe looks like a graveyard: Apple Intelligence, delayed by a year; iPhone Mirroring, still unavailable here more than a year after its release elsewhere; live translation for AirPods, pushed back; and now Siri AI indefinitely. Apple itself wrote in black and white last September that the list of features delayed in Europe would probably keep growing, and that the experience of European users would continue to fall behind.

And this is not just Apple. In July 2024, Meta simply refused to release its multimodal Llama models in the EU, citing an "unpredictable" regulatory environment — to the point that even open-source models were excluded for European companies. The refrain is the same everywhere: too much legal risk, too much ambiguity, disable Europe. We were sold the "Brussels effect," the idea that the whole world would eventually adopt our standards because we regulated first. The actual result is the opposite: the world routes around us. We no longer set the standard. We become the checkbox that says "except EU."

Ask yourself honestly. Which mainstream AI platform is European? OpenAI, Anthropic, Google, Apple, Meta: American. DeepSeek, Qwen: Chinese. On the European side, we have Mistral — which is excellent, and which I love citing — and behind it, many, many regulators. The continent that invented the Web at CERN is now specializing in the art of slowing down everyone else's innovation. We no longer build the future. We make it compliant.

"But Apple is protecting itself!" — yes, and so what?

Let us be intellectually honest, because I hate one-sided demonstrations. The opposing argument exists, and it is not stupid. Margrethe Vestager summarized it in 2024: it is quite fascinating, she said, that Apple rolls out its AI features faster where it is not required to open them to competition. Translation: "security" could be a convenient pretext to lock down the ecosystem and protect Siri's monopoly. And yes, of course Apple's economic interest is aligned with its position. A company that controls the hardware, the OS, the services, and soon the AI layer has no desire to open the floodgates. I am not going to pretend otherwise.

But that argument crashes into three walls.

The first: a choice you cannot use is not a choice. The DMA is supposed to serve consumers. The European consumer will pay exactly the same price for an iPhone as an American — around €1,200 — for a product that has been deliberately amputated. Pro-competition rhetoric, anti-consumer reality. We won the theoretical right to install a third-party assistant, and lost the assistant altogether.

The second: the security risk is not FUD invented by Apple's marketing team. Prompt injection attacks, hijacked AI agents exfiltrating passwords or photos, or modifying files without consent — these are demonstrations published by researchers, not fantasies. The more power you give an agent, the larger the attack surface becomes. Apple's architecture — on-device processing, Private Cloud Compute — exists precisely to contain that. Demanding that it be torn open in favor of "open" access for anyone is confusing openness with recklessness.

The third, and it is the most damning: Apple proposed solutions. The Trusted System Agent, the eighteen-month rollout. A company that simply wanted to protect its turf behind an excuse would not bother designing a secure interoperability mechanism and proposing a timeline. The Commission, for its part, proposed nothing in return. When only one side comes with a plan and the other replies "no" without an alternative, you know which side is actually looking for a way out.

Brussels' reply: "Apple just has to follow the rules"

The next day, the Commission responded through its spokesperson Thomas Regnier, and one thing must be acknowledged: the argument is well rehearsed. Brussels confirmed that discussions with Apple had indeed taken place upstream, and immediately sent the ball back to Cupertino. The DMA, we are told, does not prevent the launch of a new product; what it prevents is locking down an ecosystem. So if Siri AI does not come, that is Apple's choice, full stop. The Commission insisted that it is neither Apple's role to decide who has the right to innovate in Europe, nor Apple's role to dictate which AI tools citizens are allowed to use. Regnier even delivered the punchline: European law is not negotiable, just as a police officer does not ignore a speeding violation to please the driver.

It is clever. It is also, in my view, an almost complete inversion of reality.

"It is Apple's choice." Technically, yes. Except the DMA does not say, "you are not allowed to launch Siri AI." It says: "you are allowed to launch it, but the second you do, you must give any third-party assistant the same access to the user's data and apps — otherwise you are in breach, and you risk a fine, or even a sales ban." Presenting that as Apple's free choice is like telling someone they are perfectly free to speak, while forgetting to mention that the next sentence sends them to prison. The choice exists on paper. In real life, it is a constraint disguised as freedom.

"It is not Apple's role to dictate which tools Europeans use." Beautiful line. Concrete result: Europeans will use no Siri AI on iPhone, because it is precisely Brussels — not Apple — that has just decided they will not get it. In the name of freedom of choice, the only choice on the table is removed. We defend competition between assistants so fiercely that we end up allowing none of them.

On the eighteen months, the sleight of hand is even prettier. Apple proposed a gradual rollout of its Trusted System Agent, the time needed to build infrastructure capable of opening access to third-party AIs without blowing everything up. Brussels reframed that as "Apple is asking for an eighteen-month exemption from its obligations." But a staged security rollout is not an exemption: it is the normal way to put critical infrastructure into service without breaking what works. Demanding the immediate big bang, opening everything to everyone at once, is not rigor — it is a guarantee of an accident.

The most serious argument remains this one: granting that delay, says the Commission, would amount to giving Siri AI a de facto monopoly, and indirectly to Google, whose technology would partly power it. And here, let us be precise, because this is exactly the kind of shortcut that derails the entire debate. Three unrelated things are being mixed together. The first is the genealogy of the models: yes, the new Apple Foundation Models were refined through distillation from Gemini models. But that is an engineering question, not a privacy question — how a model was trained tells you strictly nothing about what it does with your data once it is in production. Confusing the two is like confusing the cake recipe with your guest list.

The second is where the largest model runs: publicly, Apple describes its server model as running inside Private Cloud Compute, on Apple silicon servers. Press reports mention possible support from Google Cloud and Nvidia GPUs for some AI components, but this point is not documented by Apple with the same level of certainty. Private Cloud Compute is end-to-end encrypted and — this is the decisive point — auditable by independent security researchers: the software images that run on those servers can be inspected and verified from the outside, by anyone. In other words, even if Google hardware or infrastructure were involved somewhere, the computer should not see your data, and nobody can claim otherwise without proving it publicly. Federighi hammered it home in a technical session: no Google Assistant, no embedded Gemini app, no call to Google's consumer models, nothing being sent to Google. The third thing — what your phone actually does with your messages and photos — therefore remains, end to end, under Apple's guarantees, verifiable and encrypted. The "indirectly Google" argument does not hold: it dresses up a question of provenance and hosting as a privacy problem, when privacy is precisely the one thing that does not move.

As for the monopoly itself, the argument collapses for an even simpler reason. Because the Trusted System Agent was exactly the mechanism meant to give competing assistants the same access as Siri. Apple was not asking to exclude rivals: it was asking for time to let them in securely. It is being accused of a monopoly it was proposing to dismantle — more carefully.

At bottom, the two sides tell the story in reverse. Apple says: "we proposed, they rejected everything." Brussels says: "we discussed, they asked for a derogation." Both can be true at the same time — there were discussions, and the Commission did reject everything. But only one question really matters: at the end of the story, who ends up without an intelligent assistant in their pocket? Us. And that is the only thing the European user will remember.

The Stone Age, limited European edition

I do advocacy with the Commission and Parliament on the regulation of Apple platforms. I know those rooms, I know the people, and I am not anti-regulation on principle — there are legitimate battles, on the App Store, steering, and fees. But this interpretation of the DMA is not regulation. It is self-mutilation. We are not protecting the European citizen, we are depriving them. We are not stimulating local innovation, we are pushing it toward the exit. We are not taming American giants, we are giving Asian and American developers a head start we will not catch up with, because in AI, advantage compounds: every month you are not allowed to learn is a month during which others learn in your place.

GDPR gave us the hell of cookie banners. The AI Act is already giving us models "not available in your region." The DMA now gives us an iPhone less intelligent than the Mac sitting next to it. By trying to write the rules before everyone else, we have written ourselves out of the game. Meanwhile, capital, talent, and product launches flow to markets where shipping is still allowed. Europe is becoming a magnificent museum: people love visiting it, life here is good, but the future is no longer built here. The Stone Age, except nicely stamped "compliant."

So let us be clear: I do not want a continent where we choose between security and innovation. I want both, and Apple was proposing a path toward both. Brussels refused to take it. At some point, we will have to wake up, stop mistaking the posture of being the world's strictest regulator for an industrial strategy, and seriously ask ourselves what we want to be: the people who build, or the people who forbid.

I chose my side a long time ago. I want to build. I want my community to build. I want us to create things here that the rest of the world looks at with envy — not watch the rest of the world through the window.

And if you have an exceptional project, contact me. Innovation is what drives me.