Before It Becomes Irreparable

The law usually arrives late. It’s its nature, almost its dignity: it looks at accomplished facts, weighs them, judges them. It works in the past tense. A ruling is always an autopsy: careful, reasoned, performed on a body that no longer breathes. That’s why it’s striking when the law tries, once in a while, to conjugate itself in the future: to stop something before it happens, because it has understood that afterward there will be no fixing it.

That’s exactly what happened on June 9, when the European Commission ordered Meta to give rival AI assistants their WhatsApp access back within five working days. The story, reported as one more Brussels-versus-Big-Tech round, hides something more interesting than a threatened fine. It hides a meditation on time: on the point where the time of markets and the time of the law stop running at the same speed, and the gap itself becomes a question of justice.

A Door That Closes in Silence

Let’s start with the facts, because they’re subtler than the telling.

On October 15, 2025, Meta updates the terms of its WhatsApp Business Solution. A line of legal text, the kind no one reads and which, for that very reason, decides far more than a press release ever does. The effect, though, is clean: from January 15, 2026, only one conversational assistant remains available on the platform, Meta AI. Everyone else, out. Not with an announcement, not with a battle: with a clause.

The point worth fixing right away is the one the fast coverage tends to skip. Meta’s leverage isn’t its model. It’s its door. WhatsApp has more than two billion users, and, for anyone building assistants aimed at businesses, it’s a distribution channel no other messaging platform can replicate at the same scale. The value isn’t in the assistant’s intelligence. It’s in the fact that the assistant can reach the two billion, in the chat the user opens fifty times a day without thinking. Closing that door to the others isn’t a contractual detail. It’s the act itself.

Freedom Denied on the Invoice

Then comes the elegant move, the one the fast coverage almost ignored. Under the pressure of the first Statement of Objections, Meta announces it will reopen to competitors. But for a fee. And the rate is set at a level that makes it economically unsustainable for anyone. Access, formally, exists. In practice, the door stays shut for everyone but the owner of the house.

It’s the old trick of the “open” sign hung on an impassable threshold: freedom granted in theory and denied on the invoice. Economists have a technical name for this figure (margin squeeze, constructive refusal to deal), but the mechanism is older than the vocabulary that describes it. You don’t forbid entry: you set the price of the ticket at a number no one can pay, and then you spread your arms and say that the door, as you can plainly see, is open. The difference between a ban and a prohibitive fee is all in the legal form; in its effect on the market there is no difference at all. It’s actually a shrewder version of the ban, because it shows up dressed as a concession.

This is the detail that separates a product decision from an exclusion strategy. A price that genuinely reopens the door is a price someone, under normal conditions, could afford. A price calibrated so that no one pays it isn’t a price: it’s a lock disguised as a rate card.

The War Has Moved Down a Layer

For years we told ourselves the AI race was a race for the best model: more parameters, higher benchmarks, the leaderboard of the month. It was a useful distraction, and probably not entirely accidental. While we watched the model layer, which is commoditizing at an astonishing pace, with frontiers compressing week by week and the gap between first and fifth now measured in months and not eras, the real game slipped beneath, to the distribution layer.

Models converge. Doors don’t.

Whoever controls the door decides who passes. And an AI assistant, however brilliant, that can’t reach the user where the user already lives is an excellence confined to a room with no exit. The recent history of the internet is a long collection of superior technologies defeated by inferior distribution: the better product losing to the preinstalled one, to the one already open, to the one that was there when the user looked. Meta knows this better than anyone: it built its empire by occupying the doors, not by winning beauty contests of technology.

Excluding rivals from WhatsApp is therefore not just any anticompetitive whim. It’s the attempt, perfectly rational from the company’s standpoint, to turn an advantage in a mature market (messaging) into an advantage in a market being born right now, that of general-purpose assistants. Economists call it leveraging: using your weight where you’re dominant to move the needle where you’re not yet. It’s a movement as natural as gravity. And like gravity, if it meets no equal and opposite force, it wins on its own. Not because it’s wicked, but because falling is what it does.

The Harm Money Doesn’t Repair

Now to the heart of it. Why an interim measure, and not the usual investigation that ends three or four years later with a fine?

The tool the Commission used, Article 8 of Regulation 1/2003, is almost a living fossil. It’s the second time in over twenty years that Brussels has pulled it from the cupboard; the modern precedent is the 2019 Broadcom case, and before that you have to go back to a time when no one was talking about artificial intelligence. Interim measures in antitrust require two things at once. An apparent infringement, the fumus boni iuris, meaning enough indications that an infringement exists. And a qualified urgency, the risk of serious and irreparable harm to competition, the periculum in mora. It’s that word, irreparable, that holds the whole philosophy of the case.

Irreparable harm is harm that compensation doesn’t heal. If the Commission were to condemn Meta tomorrow, four years from now, to pay 10% of global turnover (on $187 billion, a figure that takes the breath away), that sanction would give nothing back. Because by then the AI-assistant market will have set. Users will have grown used to Meta AI, network effects will have done their silent work, competitors will have exited or shrunk into irrelevance, and no check, however enormous, will return users to choices they were never allowed to make. Competition, once lost this way, can’t be bought back. You can fine the culprit, but you can’t resurrect the market.

This is where the law changes tense. The reasoning the Commission gave for its decision is almost a thesis in the philosophy of law: in fast-moving markets, competition can be lost long before a final decision arrives. Translated: the justice that judges accomplished facts would, here, arrive once the deed is done. Useless, and therefore unjust. So the law tries, for once, to anticipate, to behave like an ex ante regulator instead of an ex post judge.

There’s a slightly tragic beauty in this. Competition law is born slow on principle, because slowness is its guarantee of fairness: you decide after hearing everything, weighing everything, giving each side the time to defend itself. Slowness isn’t a flaw in the process, it is the process. But faced with the lock-in and path dependence of digital markets, that same slowness becomes a form of injustice: it impeccably guarantees the rights of the accused, and in doing so lets the very thing it should protect die. The speed of the door beats the speed of the robe. The interim measure is the law that, recognizing its own slowness, tries to run. It’s an exception that confesses a structural difficulty, and confessions, in law, count for more than declarations.

The Italian Detail Nobody Tells

And now the part closest to my heart, because it’s also the one most forgotten by the international coverage: none of this started in Brussels. It started in Italy.

It was the AGCM, the Italian competition authority, that moved first. Interim proceeding opened on November 26, 2025, days after the terms update, when in Brussels the case was still a file to be opened. And the Italian authority did something the Commission, in its documents, carefully avoided: it named the victims. Microsoft’s Copilot, OpenAI’s ChatGPT, Perplexity, Factoría Elcano’s Luzia. First and last names, not “third-party operators.” And it acted to protect a precise audience (the thirty-seven million Italian WhatsApp users), imposing its own interim measures as early as December, while the European level was still clearing its throat.

That move left a curious mark on the European proceeding: the Commission’s Statement of Objections covers the entire European Economic Area except Italy, where the game had already been played. A geographic carve-out that avoids redoing what was already done. And, read backward, it’s the implicit acknowledgment that the national authority had seen it right, and first.

There’s a small lesson in institutional architecture in this sequence, worth more than many declarations about European digital sovereignty. The national authority played the skirmisher: light, fast, able to strike where it knows the ground and the names of the people who live on it. Brussels brought up the heavy cavalry, extending the intervention to the continent and turning an Italian case into a European precedent. It’s subsidiarity that actually works, not as a conference slogan but as a division of labor between the one close to the problem and the one with the strength to make it precedent. For once, the machine engaged in the right order: first the one who sees, then the one who weighs.

Two Clocks, and a Third That Ticks

It’s worth keeping in mind a distinction the press happily blurs. This is not a DMA case. Meta is already a designated gatekeeper, already sanctioned on the Digital Markets Act front, but that track runs elsewhere. Here we’re in classic antitrust, Article 102 of the Treaty: abuse of a dominant position, the oldest matter there is, the one that has paraded Microsoft, Intel, Google and now Meta.

The interesting fact is that, through the interim measure, ex post antitrust ends up behaving almost like ex ante regulation. It imposes conduct before the final finding. The two clocks, the DMA’s preventive one that sets the rules before the fact and antitrust’s subsequent one that judges after, overlap on the same dial. And above both ticks a third, the AI Act, which will soon bring its grid of obligations into the very same market space. Three regimes, three times, one company.

For anyone working on European compliance, it’s the signal that the era of regulatory silos is over. You can no longer reason in compartments (the DMA in one room, antitrust in another, the AI Act in a third), a file per office and no one talking to the office next door. The boundaries between subjects are growing porous, and a company’s exposure will have to be read across the board, through regimes that speak about the same thing in different grammars and sometimes with conclusions that have to be checked against one another. Complexity isn’t a bug in the European system. It has become its language, and anyone who wants to understand it has to stop looking for the right room and start reading the corridors.

The Doors That Stay Closed

There’s a word, in this whole story, that keeps coming back: irreparable. It’s a word about time before it’s about money. It says there are harms you can’t stitch back, because they belong to a timeline that has already moved past. Some doors, once closed, stay closed. Not because someone holds the key, but because on the other side, in the meantime, the room has emptied and no one remembers there was a way out anymore. The harm isn’t the lock. It’s the forgetting that grows behind the door while it stays shut.

With this rare measure, the Commission said something simple: we won’t wait to find out whether we were right, because finding out too late would be the same as having been wrong. It’s an act of humility disguised as an act of strength: the recognition that, against the speed of platforms, the patience of the law is no longer enough, and that prudence, past a certain point, becomes complicity.

Whether it holds remains to be seen. Meta will appeal, and the appeal will test exactly this: whether a provisional measure can survive a judge’s scrutiny while the substantive investigation is still open. That is, whether the law can really afford to run without betraying the guarantees that make it law. If it holds, the precedent is heavy: dominant platforms won’t be able to exclude AI rivals from their services during an investigation, not even before a formal finding of wrongdoing. The door will stay open by law while we argue over whether it should be.

But beyond the technical outcome, the stakes are a question that concerns all of us building software in this season. Do we want the artificial-intelligence market decided by the best assistants, or by whoever owns the doors? Because the answer, if no one forces it, we already know. And when we know it for certain, it will be, precisely, irreparable.

The author

Andrea MargiovanniAI

I follow the relationship between AI and European regulation as a political fact, not a technical spectacle. I work with teams that have to make AI compliant with AI Act, CRA, NIS2 without reducing compliance to a checklist.

See the guide →