Europe has a curious reflex when confronted with new technology. Instead of first building it, experimenting with it, watching people misuse it in ways nobody predicted and gradually learning where the real dangers and opportunities lie, we seem irresistibly attracted to the opposite sequence: imagine the future, classify its possible problems, negotiate the definitions, construct the framework and then hand the result down to the people who will actually have to make it work. It is an admirably orderly approach to a world that has repeatedly demonstrated that it has no intention whatsoever of being orderly.
Artificial intelligence has turned this habit into something approaching satire. Europe managed to produce the world’s first comprehensive AI regulatory framework while most of the frontier models, the hyperscale computing infrastructure, the giant technology companies and much of the risk capital behind them were being built somewhere else. Only afterwards did the conversation acquire a distinctly more urgent tone about AI factories, gigafactories, strategic autonomy and the hundreds of billions Europe might need if it wanted to participate in the revolution rather than merely supervise it. The sequence has a certain European elegance: first decide how the game should be played, then look around the stadium and wonder why so many of the players arrived from abroad.
None of this means that AI should be left alone until something catches fire. Privacy, security, discrimination, monopoly power and the concentration of information are real concerns, and technology companies have hardly earned a historical reputation for voluntarily discovering every social consequence of their inventions. But setting broad limits is one thing; assuming that an emerging technology can be understood well enough in advance to construct a detailed administrative architecture around it is something else entirely. Nobody knows what AI will look like in five years. The companies spending billions building it do not know, researchers disagree profoundly, and investors mostly conceal their uncertainty behind impressive PowerPoint slides. Yet somewhere along the chain we develop the confidence that a sufficiently thorough regulatory process can know enough to specify how the resulting system should behave.
Once that process begins, Europe is in familiar territory. Data needs rules, AI needs rules, cybersecurity needs rules, digital platforms need rules, competition needs rules, and every one of those concerns can be defended perfectly well on its own. Unfortunately, companies do not experience laws one at a time. They experience the accumulated system, complete with overlapping definitions, reporting duties, exceptions and regulators. Eventually reality turns up, having once again failed to read the impact assessment, and inconvenient things happen: standards are not ready, provisions interact in unexpected ways, implementation dates become unrealistic and companies cannot always work out which obligation takes precedence. Then comes the uniquely European moment when legislation is introduced to simplify the legislation whose purpose was to bring clarity.
Anyone who has spent enough time in IT knows this smell. It is a legacy system being patched while everyone continues to insist that the underlying architecture is perfectly sound.
Software engineering learned, usually through expensive humiliation, that complex systems cannot be completely designed from a conference room. The old waterfall dream was wonderfully reassuring: gather the requirements, predict the exceptions, design the architecture, build everything and finally deliver a system that corresponds beautifully to the specification written several years earlier. Reality eventually cured much of the industry of that fantasy. Modern development is not perfect, but at least it begins from the humiliating recognition that users behave strangely, technologies change, assumptions fail and sometimes the clever design simply does not work. You build, observe, learn and adapt, occasionally committing the organisational heresy of throwing something away.
European policymaking still has a weakness for waterfall thinking, except that the project plan is written into law.
The continuing quarrel with Apple is a wonderful example because neither side fits comfortably into the usual morality play. Apple has spent years constructing an unusually integrated environment in which hardware, operating system, applications, payments, identity and security mechanisms are designed as parts of one system. That integration is also commercially magnificent for Apple: it creates lock-in, protects lucrative services and makes life considerably harder for competitors. The EU therefore has legitimate competition questions to ask. But the same closed architecture also produces genuine benefits in privacy, security and predictability. It is not merely a monopoly trick accidentally decorated with encryption.
The DMA (Digital Markets Act) looks at such a system from a different direction. If a platform has become powerful enough, competitors must be able to obtain access to capabilities that the platform owner reserves for itself. Translated from the legal language into something slightly more mischievous, Europe can sometimes sound as though it is saying: very nice ecosystem you have built there, very safe and elegant, but unfortunately it does not quite conform to our model of how an ecosystem should be arranged.
Siri AI has made this clash wonderfully concrete. Apple released its new Siri AI in September 2026; ; EU users can use it on supported Macs and Vision Pro devices but it remains unavailable on their iPhones and iPads. Apple says the DMA interoperability obligations would require access for competing assistants in ways it has not managed to reconcile with its privacy and security design. The Commission rejects the suggestion that the DMA prevents Siri AI from launching and says Apple remains free to introduce it provided competitors receive equivalent access, subject to user consent. There is an intriguing detail underneath that argument: iOS and iPadOS are designated core platform services under the DMA; macOS is not. The result is that the European owner of a MacBook and an iPhone can sit at the same desk, under the same European laws, and receive the new Siri AI on one Apple device but not the other.
You could hardly invent a better illustration of the problem. Apple may indeed be defending its moat, and Brussels may indeed be defending competition; both propositions can comfortably be true at the same time. What deserves more scrutiny is the confidence behind the regulatory assumption that an integrated technical architecture can simply be opened at selected points while all the desirable properties of the original system—privacy, security, reliability, competition and user control—remain available because the regulation says they should. Engineers tend to become nervous at this point because complex systems are famous for connections that become visible only after somebody touches them. Bureaucrats have the luxury of writing several desirable outcomes into the same document and leaving the engineering department to discover whether they can coexist.
If they cannot, presumably there will be guidance.
That is the deeper battleship. Europe’s problem is not merely that it regulates too much; it is that it has too much faith in regulation as an instrument of foresight and control. Faced with uncertainty, the instinct is to reduce uncertainty administratively, even when the uncertainty exists because nobody yet knows enough. This becomes particularly awkward when Europe is already technologically behind. Apple, Google and Microsoft can hire another regiment of lawyers and engineers to deal with Brussels. A young European company cannot. Complicated rules may annoy an incumbent while quietly helping to ensure that fewer challengers ever become incumbents.
Europe talks increasingly about technological sovereignty, but sovereignty cannot be created through regulatory vocabulary. If Europe wants influence over AI, cloud computing, semiconductors and whatever comes next, Europeans eventually have to build important parts of those industries. A continent that depends heavily on technology produced elsewhere while becoming exceptionally sophisticated at determining the conditions under which that technology may be used certainly wields regulatory power. Whether that amounts to technological power is a rather different question.
The alternative is not a regulatory Wild West. It is something much less dramatic and considerably harder for institutions accustomed to control: humility. Establish firm boundaries where the danger is clear, but accept that uncertainty cannot be legislated away. Allow experiments, measure what actually happens, correct quickly when assumptions fail and make it normal for rules themselves to disappear when experience proves them unnecessary or misguided. Regulation, like technology, needs feedback loops.
The rebellious IT guy would probably reduce the entire argument to one sentence: if you have never run the system in production, don’t pretend you already understand every failure mode.
Europe does not need to choose between innovation and protection. It needs to stop believing that protecting people requires first predicting the future and then forcing that future into a model designed in advance. Sometimes the wiser approach is to build carefully, watch closely and learn quickly, because while Europe is still perfecting the specification, somebody elsewhere has usually discovered that the specification describes last year’s world.