The Spectre We Are

The Spectre Is Us

A spectre is haunting Europe, and for once it is not the one Marx had in mind. It is a stranger spectre, because it does not haunt those of us who live here. It haunts those who watch us from the outside, from the other side of the ocean, who look at us the way one looks at an eccentric relative at Christmas dinner—someone you are obliged to put up with, someone you would happily be rid of, who nonetheless stubbornly keeps existing and keeps talking about strange things like the dignity of work, accessibility, child protection, product liability. The spectre is us. Or rather, it is what we are doing here in the area of digital regulation, a series of acronyms that in Brussels are considered ordinary technical work and that in San Francisco are perceived as a near-existential threat.

Seen from the outside, and in particular seen from one specific corner of the world that has made itself heard a great deal in the last few years, Europe has become a character. It is no longer an economic subject, nor a political union, nor, by now, the continent of the hundreds of millions of people who actually live here. It is a meme. A meme with a precise narrative function: to embody what the future must not be. Nineteenth-century bureaucracy pasted on top of the twenty-first century. The old world holding back the new. The open-air museum that presumes to dictate rules to tomorrow’s factory. It is a caricature, of course, but an effective one, because it works inside a much larger story than itself, and that story is the culture war a slice of American capitalism is waging against the very idea that software can be the object of rules.

GDPR and the Banner That Sums It Up

To understand where we are now, it helps to go back a little. The moment when the relationship between Silicon Valley and European regulation cracked in a way that has been hard to mend was GDPR. A regulation from 2016, applied from 2018, technically not revolutionary, drawing largely on principles that already existed in Convention 108 and in directive 95/46. Yet what GDPR did in practice was to declare that certain business models built on indiscriminate collection of personal data would, at some point, have to answer for something. It is no accident that GDPR has become the symbolic target par excellence in the anti-European narrative, and no accident that the prevailing way to attack it is the cookie banner.

Here I owe my first concession. The cookie banner as we live with it is a failure. It is a failure from the user’s point of view, who clicks it away without reading. It is a failure from the standpoint of effective protection, because that consent is almost always fictitious. It is a failure from the standpoint of the interface designer, forced to accept that every European site opens with a digital customs gate that nobody wants. It is the textbook example of a rule that is correct on the level of principles and that has discharged itself onto the implementation plane in a grotesque way, because the legislator assumed consent was the sovereign act of will of a rational subject, and the designers of digital products knew perfectly well that you only had to design the banner a certain way to get it clicked through in half a second. The dark pattern has eaten the theory of free, specific, informed and unambiguous consent. You can see it every day. It is a real problem. The critics are right on this specific point.

But this is where the divergence between useful and instrumental criticism begins, because anyone who is right about the cookie banner is not actually arguing about the cookie banner. They are using the cookie banner as a metonymy to attack the very idea that the collection of personal data is a regulable problem. They are saying, in substance, that if the rule produces an ugly implementation then the rule must be abolished, not that the implementation must be corrected. It is an old rhetorical move, and there is no need to summon any authority here—it is enough to observe that those who call for the abolition of GDPR on the basis of the cookie banner do not usually call for the modification of the technical consent rules, which would be the coherent position. They call for a return to a world in which American platforms can do whatever they want with our data without owing anyone an explanation. The banner was a pretext, not an argument.

DSA, DMA, and the Business Model

That it was a pretext became definitively clear with the DSA and the DMA. When the Commission began asking gatekeepers to open their platforms, to make certain services interoperable, to stop favouring their own products in their stores, to allow users to uninstall pre-installed applications, the argument “you regulate too much” turned without intermission into the argument “you are attacking our companies”. The leap is interesting. For years they explained to us that the problem was the excess of rules, not the identity of those who paid. Once the rules became operational, they explained that the problem was that they hit only American champions. The two theses are not compatible, but they coexist comfortably inside the same line of argument, and that alone should make one suspect that something does not add up.

A point worth making about the DMA. The thesis that the DMA specifically targets US firms is factually weak, because ByteDance is also among the gatekeepers, and because thresholds are thresholds. A European company reaching seventy-five billion in capitalisation and forty-five million monthly end users would be a gatekeeper too. The problem is that European companies in that range are extremely rare, and the only recent case, Booking, operates from Amsterdam but reports to a US holding. That absence does not depend on the DMA. It depends on industrial, fiscal, financial and educational choices of the last forty years that the DMA did not cause and will not solve. If we want to talk about that problem, let us talk about it seriously, citing the Draghi report if we need a recent anchor, but let us not confuse it with platform regulation, because the two planes only touch in appearance.

This is the point that those who watch us from outside do not want to see, or perhaps see perfectly well and pretend not to. Platform regulation is not in competition with innovation. It is in competition with a specific business model—mass extraction of personal data, attention optimisation pushed to the edge of addiction, capture of adjacent markets through dominant-position leverage, systematic offloading of negative externalities onto the social body. It happens that this business model has been the prevailing model of American digital capitalism for the last twenty years. It happens that the companies practising it are now the largest in the world, and that they have an obvious interest in continuing to practise it. If the European Union says that this specific model has limits, it is touching a raw nerve, and not by accident. It is saying, more or less articulately, that there is another way of doing digital technology. In 2026, that is heresy.

From Rule to Technical Object

Let me leave analysis aside for a moment and tell a concrete story, because I think it helps to know where I am speaking from. I have worked for several years inside a mid-sized Italian ICT company, outside the major metropolitan areas, that builds software for public administration and healthcare. For the last two years my work has been largely one thing: translating European regulation into technical objects. What it means to have a healthcare platform compliant with GDPR Article 28 when the supplier of a supplier is in Germany and the health data concerns an Italian citizen. What it means, for a training body delivering workplace-safety courses under the Italian Accordo Stato-Regioni, to see the Cyber Resilience Act arrive and realise that its LMS falls within scope. What it means, for a software house producing a back-office application, to prepare for the fact that the revised Product Liability Directive has extended the regime of defective-product liability to software—including certain forms of integrated AI components—and that the regime will apply operationally to products placed on the market after December 2026. What it means to design a DPIA following the template approved by the EDPB last March, only to realise that it is not a form but a documentary genre, to be kept alive alongside the system it describes. These are jobs of translation, not of abstraction. The rules that on the other side of the ocean are painted as obstacles to innovation are, for me, the innovation itself, because they define the perimeter inside which one can design well.

I do not say this out of romantic adherence to the European project. I say it because, after implementing these requirements dozens of times inside real systems, you see something an Andreessen op-ed cannot. You see that conformity is not an isolated cost, it is a design pressure. Like all design pressures, it narrows the space of choices and in exchange forces you to do things better, at least within the dimension the rule protects. The PLD, for instance, shifts producer liability for software quality in a way that until yesterday did not exist even as a category. For those who sell serious products this is an administrative nuisance. For those who sold junk while promising it was magic, it is an earthquake. Guess which of the two groups produces the most virulent criticism.

One could say, and people do say, that this argument holds for my small Italian shop but not for the technological frontier. The Italian shop is not inventing foundation models, not designing gigawatt data centres, not redefining the paradigm of scientific research. True. But that is precisely why my vantage point seems useful to me, because it tells from very close range what the great Californian narratives never manage to see, namely that the vast majority of software running in the world is not frontier, it is everyday infrastructure. It is the healthcare systems handling our clinical data, the public-administration portals handling our relationship with the state, the back-office systems running small businesses, the training systems that certify legally mandated competencies. It is fabric. European regulation is written above all for this fabric, and the fabric, by definition, needs rules in order to exist as fabric and not as a random pile of threads.

AI Act, Vance, and the Race

Let me move on, because there is another theme that deserves to be taken seriously, and that is the AI Act. Here too the concession must be made. The AI Act is a difficult regulation, written at a historical moment when the technology it was meant to regulate was changing every four weeks. It had a complicated gestation, took a mid-course turn when foundation models arrived, and produced in the final text certain asymmetries between obligations on providers and obligations on deployers that operators are still trying to figure out how to apply. There are ambiguities in the codes of conduct on general-purpose systems. There are doubts about how to compute compute-thresholds. There are operational difficulties in conformity assessment for high-risk applications in healthcare, and here I speak from direct experience because Trustie, the platform I work on as CTO-as-a-Service for Umana Analytics, has to wrestle with some of these doubts alongside GDPR and other pieces of European discipline that overlap in not always linear ways.

All true. All fair. And it is precisely for that reason that the AI Act is not the end of the world. It is a young rule, in its bedding-in phase, dealing with a sector that is itself young and in its bedding-in phase. The application difficulties are reciprocal, because the technology is in the phase of discovering its own space and the rule is in the phase of discovering its own implementation. This is normal. The same has happened with every transformative technology—industrial chemistry in the early twentieth century, nuclear safety in the second half, medical devices in the 1980s, finance after Lehman. New rules live their first years in a state of iteration. The difference today is that American operators consider it normal for the law to chase technology, while they consider it scandalous for the law to stand up first.

The scandal is told in almost religious terms. It is not regulation, it is hyper-regulation. It is not caution, it is cowardice. It is not public interest, it is socialism in legal disguise. Vance, in Paris, last year, gave a speech that will go in the textbooks not for its content but for its tone. He explained, to the European audience, that we are strangling AI with our fears, that we are prisoners of our own due-process scruples, that the future will not wait. It was a speech designed to be shared in forty-second clips. It worked very well for that purpose. It worked less well as an analysis of a rule that, let us recall, does not ban the development of AI but articulates it according to risk levels. The conceptual framework of the AI Act is the same framework with which any industrial civilisation regulates dangerous chemical substances, pharmaceuticals, automobiles, food intended for children. No one is asking Vance to stop thinking that regulation is always excessive—it is a legitimate political position with its own tradition. Only, let him present it for what it is, not as the neutral observation of a neutral observer.

There is an argument one hears often that I think deserves a reply. It is the argument of the race. AI is a race, we are told, and Europe is losing it. While we write regulations, China builds, the United States builds, and in five years we will be irrelevant. Fine. I concede this too, in part. On computational infrastructure and venture capital availability Europe is in fact behind, and will be for a long time, and this is a serious problem that would require real industrial policy, substantial public investment, integration of the European capital market, reform of academic labour, none of which we have been seriously discussing for decades. But the race, what race exactly. A race is defined by its finish line. What is the finish line. Having our own version of OpenAI. And then. Having our own version of Meta. And then. That is, having companies that do what, for whom, to obtain what, selling what to whom. The question is not idle, because the companies that Vance’s friends propose to us as a model are companies whose value is largely built on a precise assumption: that personal data can be collected and used with minimal limits and that attention-based business models can scale without substantive constraints on child protection. If Europe wants to be the one chasing that model and trying to replicate it, then yes, it is losing. If Europe wants to be the one proposing an alternative, the race is a different race. It plays on different ground. It is won by different rules. And in that case regulation is not a brake. It is the product.

Brussels Effect and the Stars-and-Stripes Regulatory Wall

I realise this sentence sounds presumptuous, and in part it is. But it is the heart of the matter. The Brussels Effect, which Anu Bradford has been writing about for at least a decade, is not a rhetorical invention. It is a concrete mechanism. When the European Union sets a standard, and when the European market is large enough that it cannot be ignored, global firms end up adopting that standard as default, because keeping two parallel architectures costs more than keeping one. GDPR has been an informal default for years in the design of data-management systems inside companies that operate globally. European accessibility specifications end up being implemented everywhere, because anyone selling even a third of their products in Europe designs a single interface. The CRA will produce a similar effect on software. The PLD will produce a similar effect on product liability. The AI Act will produce a similar effect on the risk classification of automated systems. This is not reverse colonial imposition, it is the ordinary functioning of the market, and it is precisely for that reason that it angers the techno-optimists, because their companies are forced to adapt to rules they did not write.

Then there is the regulatory-wall argument. Europe, we are told, is building a regulatory wall that excludes American suppliers. This argument deserves to be taken seriously, because part of it is true and part of it is false, and those who use it usually mix the two. The true part is that some regulations, in particular the Data Act, certain provisions on coordinated cybersecurity, certain rules on sovereign clouds, do have an industrial-policy component, in the sense that they try to encourage the emergence of a European technological offering in critical sectors. This is no mystery, it is written in the recitals, and it is a legitimate activity that every state and every economic area in the world undertakes. China does it openly, the United States does it with the CHIPS Act and the Inflation Reduction Act, with the renewed control over foreign investment, with the export bans on advanced GPUs. The novelty is not that Europe is doing it. The novelty is that it is doing it with a European specificity, based on transparency and accountability requirements, on an explicit concern for fundamental rights. To the eyes of an American investor in the Andreessen Horowitz orbit, that is an anomaly. It is one to the extent that the American investor regards regulatory activity as, in itself, a political aberration.

The false part, instead, is the claim that the European regulatory wall is different from any other regulatory wall in the world. Try exporting medical software to the United States without the FDA. Try selling a financial service to American consumers without SEC or FINRA authorisation. Try operating an autonomous vehicle at federal level without engaging the NHTSA. Try selling toys for children without the CPSC. The American system is full of regulatory walls, those walls have been there for decades, and they work very well as barriers to entry for foreign suppliers. When a European company complains about their weight, the standard American answer is “rules are rules, if you want to sell here you adapt”. It is exactly the same answer we give today to anyone protesting the CRA, and it has exactly the same legitimacy. Only, we give it from a position of narrative-power asymmetry that places us, from the start, on the wrong side of the argument.

The Narrative Asymmetry

This narrative asymmetry is the real spectre haunting Europe. Not the Europe that frightens others. The Europe that frightens itself, because the only dominant cultural frame in the tech sector is the one elaborated in California over the last twenty-five years, and that frame considers every regulatory act a failure by definition. From inside that frame, every European rule is a symptom of decadence. Every directive is a sign of stagnation. Every attempt to articulate a different vision of the relationship between technology and citizens is dismissed as the resentment of the loser of the race. It is a very effective frame, and it is very hard to contest from inside European tech, because European tech itself has often modelled itself on American tech, reads its blogs, adopts its tools, absorbs its vocabulary, and ends up judging its own continent through someone else’s eyes.

Let me push one step further. When I say the Californian frame is hegemonic, I am not doing barstool sociology. I am describing the fact that, at any European conference on digital in 2026, a large share of the technical lexicon, of the virtuous examples, of the interpretive frame comes from American industry literature. It is a fact. From here the problem flows, because when the European regulator tries to write a rule, it has to do so in an intellectual environment populated by operators who have absorbed the adversary’s grammar. They are not always aware of it. Often they think in good faith they are being “pragmatic”, that they stand on the side of the “facts”, that they want to “avoid hyper-regulation”. But they are using a vocabulary made elsewhere, for someone else’s purposes, and they end up producing, even from European positions, an implicit defence of the model they would like to criticise. The result is that in the corridors of Brussels progressive regulations are debated while in the feeds of Italian founders Sacks’s clips circulate explaining how those regulations are an aggression against the free market. The two coexist. More than that, they feed each other. Every rule written in Brussels produces a new round of complaints in San Francisco, every round of complaints in San Francisco produces a new cultural concession from a slice of the European business class, every cultural concession produces, downstream, a political weakening of the next legislative process.

The Project, and Why It Must Be Named

At this point, after two waves of concession and one of analysis, the reader has probably guessed where I am going. I would like us to be very clear on one thing. The hatred for European digital regulation, the real, visceral kind found in David Sacks’s posts, in Joe Lonsdale’s interviews, in Balaji Srinivasan’s paranoias, in JD Vance’s set-pieces, is not a technical disagreement. It is a political position, and a precise political position, and it must be recognised for what it is. That position holds that nation states should retreat in the face of tech companies, that tech companies should enjoy a functional sovereignty equal to or greater than that of states in the territories where they operate, that representative democracy is too slow for the rhythm of innovation, and that in case of conflict it is democracy that must adapt. This is a coherent position, and a position with thinkers, books, manifestos, even refined theoretical elaborations behind it, from Ayn Rand’s libertarianism to Curtis Yarvin’s neo-reaction by way of right-wing accelerationism drawing on Nick Land, all the way to Thiel’s most recent syntheses on stagnation. It is not caricature, even if its popular version is. It is a project.

It is worth pausing on this project for a moment, because for a long time it was underestimated in Europe, and perhaps still is. The strand starting from Yarvin, in particular, openly maintains that representative democracy is a dysfunctional system and that it should be replaced by forms of government run as enterprises, by CEOs with full powers, untethered from constitutional constraints. Until ten years ago this was a position regarded as folkloristic even by mainstream American conservatism. In the last five years it has normalised, entered the conversations of Sand Hill Road, found implicit and explicit funding inside Thiel’s network, colonised parts of the Republican Party, and ended up sitting inside the White House with a vice president who quotes Yarvin in public without embarrassment. To put it another way, a significant slice of the current American ruling class regards the post-war liberal-democratic order as a failed experiment, and regards its own techno-corporate model as the natural historical successor. In that frame, the European Union is not just a commercial nuisance. It is, literally, an embodiment of everything they want to overcome: a supranational subject that imposes transparency standards on firms, that protects competition, that places limits on the producer’s freedom in the name of rights that cannot be bought or traded. For those who think the future should be governed by the best, and that the best are the winners of the market, Europe is a historical enemy.

I realise that put this way the matter may sound exaggerated, and that many Italian readers, perhaps not prejudicially hostile to American tech culture, will tell me I am caricaturing in reverse. I am not caricaturing. I am simply taking seriously the things the people in question publish openly on their own profiles. Go and read Marc Andreessen’s interviews from 2023 and 2024. Go and look at the techno-optimist manifesto. Go and read Yarvin’s 2024 book. Go and listen to the All-In podcast, especially the segments commenting on European regulation. Go and watch Vance’s speech at the Paris AI summit in February 2025. I am not reading anything between the lines. It is all above the lines, written with a frankness that, from where we stand, is almost disarming. The only thing that stays between the lines is the political conclusion, and the political conclusion goes like this: democratic authority has no standing to set limits on the activity of tech companies, because tech companies are building the future and democratic authority belongs to the past.

Against this project, the European Union, with all its slowness, all its imperfections, all its cookie banner, is in fact the principal contemporary geopolitical obstacle. It is not because it has a plan against American tech, but because its model embeds an assumption the Californian project wants to demolish: the assumption that democratic authority can legitimately set limits on the economic activity of firms. That is all. It is an old assumption, present in the constitutional tradition of every European country and of the pre-Reagan United States, but in the cultural climate of 2026 it has become heretical. Our heresy consists in taking up that assumption again and applying it to digital. Nothing more, nothing less. That is why we are a spectre: because we remind the world, by our very institutional existence, that thinking otherwise is possible.

Coherent Europe, Misplaced China

Let me pause to answer an objection I can imagine. But is Europe really that coherent. Are there no internal divisions, no lobbies, no national manoeuvres, no regulatory mess. Of course there are. Europe is a complex political system, with diverging interests among member states, with Germany often defending the interests of its automotive industry even when this conflicts with environmental coherence, with France often trying to promote national champions under the cover of European industrial policy, with Italy often weakly represented technically at the tables that matter, with national governments using hostility to European regulation as a tool for domestic consensus. All this exists. But none of it changes the fact that the regulatory output of the European process is, in the end, coherent with a certain model of the relationship between technology and citizens, and that this model is radically different from the American one, and that the difference is the real object of the conflict.

Another objection. And China. China, in the anti-European discourse, is always wheeled out as a reductio ad absurdum: you regulate, while China builds, and China will win. This is a rhetorical move so ramshackle that often it is not worth engaging with, but one detail is worth noticing. The Chinese model of digital governance is one in which the state, identified with the Party, exercises a very deep control over platform behaviour, content, data, international flows. Beijing has a level of regulatory intervention over the Chinese TikTok algorithm that would make the DSA look like a kitchen-robot instruction booklet. Yet when the techno-optimists cite China as a model of non-regulation, they do not say so. They limit themselves to citing the speed at which infrastructure is built, the speed at which AI applications spread, the speed at which sites like Temu and Shein are shut down. They skip, with notable elegance, the existence of the state. This is because, deep down, their problem is not the state as such, it is a democratic state that imposes constraints on their friends. Faced with an authoritarian state that imposes generic constraints on everyone while guaranteeing itself an unchallengeable dominant position, they have far less to object to. The point is democracy, not regulation. Regulation is just the form that democracy, in a digital world, takes in legal terms.

Technique in the Service of Politics

I would like to close by stitching some of the fabric back together, because I started by quoting Marx and it is right that the circle should close. The spectre of the Manifesto was a promise of transformation. It was something that, in the eyes of the nineteenth-century ruling classes, represented the possibility that the future would not simply be the continuation of the present. The spectre that today disturbs the dreams of Californian venture capitalists promises no revolution. It promises only that the digital future can also be designed by subjects other than their portfolios, according to principles other than theirs, inside an institutional frame that precedes their model and will survive its eventual crisis. It is a modest spectre, and that is precisely why it is unbearable, because it dismantles the claim of exclusivity that the Californian model has arrogated to itself for a quarter of a century. It dismantles the idea that there is only one legitimate way to do digital technology, according to rules written by a specific community of investors in a specific geographic area of the planet. It returns to technology its status as a plural cultural artefact, subject to plural institutions, contestable and contested. For those who believed they had won history, this is an affront.

For those of us who do this craft, on this continent, in these years, the spectre is not a nuisance. It is a compass. It is what tells us, every morning, when we open a pull request or write a tender specification, that technical choices have political consequences and that political consequences deserve to be protected by law. We do not do this because we are decadent. We do it because we have a long historical memory, and we know that technologies without limits, in the previous four hundred years, have produced the exact opposite of the freedom they promised. We do it because we have read Polanyi, we have read Hannah Arendt, we have read the constitutional fathers of our continent, we have read the European jurisprudence on fundamental rights, and from all this material we have drawn a simple conclusion. The market is not a fact of nature. It is an institution. And like all human institutions, it has to be built and corrected over time. Digital is a market. Its rules do not descend from the Silicon Valley sky. We write them too, here, in the middle of a continent that is slowly noticing it has become the strangest place in the world: the place where someone still tries to say that technique should serve politics, and not the other way around.

There is one last thing I would like to say, and I will say it in a more personal register, because that is the right register to close in. When I happen to talk about these things with American colleagues, usually within three minutes the moment arrives when I am asked, with a shade of sincere pity, whether I do not feel suffocated, in Europe, by all this bureaucracy. The question is asked in good faith. They genuinely think we live in a kind of regulatory cage that prevents us from building. The most honest answer I can give is that no, I do not feel suffocated. I feel I am in the right place. Because here, at least here, the question “what does this software do to people” is a legitimate question, it is a question that can be put to a board of directors, it is a question that has legal consequences if the answer is bad. Elsewhere the same question is considered, at best, naive. At worst, subversive. There. I prefer the state of affairs in which I can ask it, even if to ask it I have to accept a little more paperwork, a few audits, a few EDPB templates, a few conformity assessments. It is a price. You pay it. It is not a cataclysm. If anything, it is the price of civilisation.

The spectre haunting Europe is us. And we will keep on being it as long as someone, somewhere, still considers technique to be a political question and not a question of pure engineering. That is no small thing. That is no small thing at all.

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 →