If You Do Not Build the Machine, Can You Still Shape What It Becomes?

If You Do Not Build the Machine, Can You Still Shape What It Becomes?

For much of the twentieth century, Europe could plausibly imagine itself as one of the places where the technological future was made. Today, the geography of artificial intelligence looks different. The companies most strongly associated with frontier AI are overwhelmingly American, while China has developed its own powerful technological ecosystem and has made artificial intelligence a strategic priority. Europe remains scientifically and technologically significant, but it is rarely described as the center of the contemporary AI race.

Yet Europe has not disappeared from that race. It may simply be participating in a different one.

The European Union has increasingly positioned itself not only as a producer or adopter of digital technologies, but as a regulator of the conditions under which those technologies may enter markets and societies. The adoption of the EU Artificial Intelligence Act in 2024 made this role particularly visible. Rather than regulating AI primarily according to a particular technology or industrial sector, the Act organizes many of its obligations around categories of risk and the social settings in which AI systems are deployed.

This produces a question that is simultaneously technological and political:

If you do not build the machine, can you still shape what it becomes?

The question matters because technological power is often understood in terms of production. Who has the largest companies? Who possesses the most advanced models? Who controls semiconductor supply chains, computing infrastructure, data and capital?

But rules can also shape technologies.

The European case therefore invites us to consider another form of power: the ability to define the conditions under which technologies become socially acceptable, commercially viable and legally permissible.

Three ways of governing the digital world

Anu Bradford provides one influential framework for understanding this geopolitical landscape. In *Digital Empires*, she distinguishes between three broad models of digital governance: a predominantly **market-driven American model**, a **state-driven Chinese model**, and a **rights-driven European model**. These categories are ideal types rather than complete descriptions of each political system, but they identify an important difference in emphasis.

The United States has historically relied heavily on private firms, market competition and technological entrepreneurship. China has integrated technological development much more directly into state strategy. The European Union, by contrast, has built much of its digital influence through law: competition rules, privacy legislation, platform regulation and, increasingly, AI governance.

Europe's regulatory approach should not be mistaken for an absence of technological ambition. The EU continues to invest in computing infrastructure, research and industrial AI. But regulation has become one of the most visible instruments through which Europe exercises power in the global digital economy.

This is particularly important because regulation does more than prohibit certain activities after technologies have already been created. Depending on how it is implemented, regulation can influence decisions much earlier in the technological process: what data are collected, what documentation is maintained, how risks are assessed, which systems require human oversight and what kinds of applications companies decide are commercially worthwhile.

The boundary between technological design and political regulation is therefore less clear than it initially appears.

The Brussels Effect

The idea that European regulation can influence behavior outside Europe predates the AI Act.

Bradford famously described this phenomenon as the **Brussels Effect**: under certain conditions, companies operating globally may adopt European standards beyond the territory of the EU because maintaining different products or procedures for different jurisdictions can be expensive or impractical.

The logic is relatively simple.

A company based in California, Seoul or Shenzhen may not be legally European. But if the European market is sufficiently valuable, that company may modify its practices to satisfy European requirements. If maintaining a separate European version of a product becomes costly, the company may extend those practices elsewhere.

European regulation can therefore travel without another government formally adopting European law.

The AI Act contains an especially interesting version of this dynamic because its scope is not determined solely by where an AI provider is physically located. Under Article 2, certain providers or deployers established outside the Union can fall within the Act when, among other circumstances, the output of their AI systems is used within the EU.

A border therefore does not necessarily protect an AI system from European regulatory influence.

But geographical reach alone does not prove global influence.

The more difficult question is whether compliance with European rules actually changes AI systems outside Europe.

When regulation enters the design process

Consider a hypothetical company developing an AI recruitment system in the United States.

The company may initially think of European regulation as something external to engineering: a legal matter to be addressed by lawyers once the system has already been developed.

But compliance requirements can move upstream.

If the system must satisfy requirements concerning documentation, risk management, data governance, transparency or human oversight before being deployed in Europe, engineers may need to design those capabilities into the system itself.

At that point, regulation no longer operates only at the point where the product enters the European market. It can begin influencing architecture, testing procedures, documentation practices and organizational decision-making.

This is where the Brussels Effect becomes particularly interesting for artificial intelligence.

The border is no longer simply located at customs.

It can enter the development process.

A European legal requirement may ultimately appear inside a technical workflow in a company thousands of kilometers away from Brussels.

That possibility has led scholars to ask whether the AI Act might generate both a **de facto Brussels Effect**, through company behavior, and a **de jure effect**, through governments adopting similar regulatory ideas.

But the evidence is increasingly complicated.

Influence is not the same as control

Early discussion around the AI Act sometimes assumed that Europe would repeat what it had achieved with data protection regulation: regulate a large market first and encourage the rest of the world to converge around European standards.

Recent scholarship is much more cautious.

Ben Crum argues that the international influence of the AI Act may be better understood through "experimentalist governance" than through a simple Brussels Effect. Instead of the EU producing one dominant global standard that other jurisdictions copy, different governments may develop their own systems while borrowing, adapting and revising elements of the European model.

That distinction matters.

Regulatory influence does not require replication.

A government may adopt the language of risk classification while designing very different enforcement mechanisms. Another may adopt transparency requirements without reproducing Europe's fundamental-rights framework. Technical standards may gradually emerge through negotiation among regulators, companies and standards organizations rather than through unilateral European rule-setting.

Under this interpretation, the AI Act is not necessarily a blueprint for the world.

It is one experiment entering a global conversation.

An even more recent study by Sina Hoch and Daniel Mügge directly examines the prospects for a Brussels Effect in AI governance. Their analysis identifies significant constraints on the EU's ability to transform global AI governance. In particular, the structure of the AI industry differs in important respects from sectors in which European regulatory influence has previously been strongest. Frontier capabilities, computing resources and leading AI firms remain heavily concentrated outside the European Union.

This changes the balance of power.

A regulator is strongest when companies cannot easily ignore its market and when complying with separate regulatory regimes is sufficiently expensive that global standardization becomes attractive.

AI may not always satisfy those conditions.

Software can sometimes be modified across jurisdictions more easily than physical products. Large AI companies may have sufficient resources to maintain different systems, interfaces or compliance procedures in different markets. Governments outside Europe are also developing their own approaches rather than waiting for Brussels to define the rules.

Europe's influence may therefore be substantial without becoming universal.

From exporting rules to exporting vocabulary

There is, however, another kind of influence that is harder to measure.

Even where countries do not reproduce the AI Act, European regulation may help establish the vocabulary through which AI governance is discussed.

Risk.

Transparency.

Human oversight.

Fundamental rights.

High-risk applications.

General-purpose AI.

Systemic risk.

Once such concepts become embedded in international policy debates, companies, governments and researchers may begin structuring arguments around them even when they disagree with European solutions.

This might be called a weaker form of regulatory power.

Europe does not determine the final rule, but it helps determine the questions that others feel required to answer.

This distinction between "rule diffusion" and "conceptual diffusion" deserves more attention.

A jurisdiction may fail to export its legislation while successfully exporting the categories through which a technological problem becomes politically legible.

If that occurs, the EU's influence would be more subtle than the strongest versions of the Brussels Effect suggest.

Europe would not govern the world.

But it might partially shape how the world thinks about governing AI.

The paradox of regulatory power

This creates a paradox.

The EU's regulatory strength partly derives from the size and attractiveness of its market. Companies accept European requirements because they want access to European consumers, institutions and businesses.

But the long-term credibility of this power may also depend on Europe remaining technologically and economically important.

If Europe becomes increasingly dependent on technologies designed, trained and controlled elsewhere, its regulatory autonomy could weaken.

A regulator can set conditions for entry into its market.

But what happens when the technologies entering that market become infrastructural necessities?

What happens if European industries become deeply dependent on a small number of non-European AI providers?

At what point does the relationship reverse?

Instead of technology adapting to European regulation, European regulation may begin adapting to technological realities created elsewhere.

The question therefore should not be whether Europe is “winning” or “losing” the AI race. Such language compresses several different forms of power into a single competition.

Technological capacity is one form of power.

Capital is another.

Infrastructure is another.

Standard-setting is another.

Regulatory legitimacy is another.

Europe's position becomes clearer when these dimensions are separated.

The European Union may possess less frontier AI capacity than the United States or China while simultaneously exercising considerable influence over how AI is governed.

Both statements can be true.

Shaping a machine you did not build

The deeper question is whether regulation can shape technology itself rather than merely constrain its use.

There are reasons to think it can.

Engineering systems are never developed in isolation from institutions. Design decisions respond to budgets, liability, procurement rules, technical standards, organizational incentives and legal requirements.

A requirement for explainability can affect model choice.

A requirement for documentation can affect development procedures.

A prohibition on a particular application can redirect investment.

An obligation to maintain human oversight can influence interface design.

In this sense, law can become one of many inputs into engineering.

But there are equally strong reasons for caution.

Regulatory ambition does not automatically produce regulatory capacity. Enforcement matters. Technical expertise matters. International interoperability matters. Market leverage matters. And regulation written today may struggle to govern systems whose capabilities change far more quickly than legislative institutions do.

For that reason, the most interesting interpretation of Europe's AI role may lie somewhere between triumph and irrelevance.

The EU is neither simply standing outside the AI revolution with a rulebook, nor clearly writing the operating system for global AI governance.

It is conducting a large political experiment.

The experiment asks whether a jurisdiction that does not dominate the production of a technology can nevertheless influence its trajectory by defining the social and legal conditions under which that technology may be used.

The answer is still uncertain.

Perhaps the Brussels Effect will extend deeply into AI development.

Perhaps European concepts will travel while European rules do not.

Perhaps companies will comply inside Europe while continuing entirely different practices elsewhere.

Or perhaps the AI Act will become only one component of a more plural global regulatory system.

But the uncertainty itself reveals something important.

The struggle over artificial intelligence is not only a struggle over who builds the most powerful machine.

It is also a struggle over who gets to decide what the machine is allowed to become.

And in that second struggle, Europe is far from absent.

References

Bradford, A. (2023). *Digital Empires: The Global Battle to Regulate Technology*. Oxford University Press.

Crum, B. (2025). “Brussels effect or experimentalism? The EU AI Act and global standard-setting.” *Internet Policy Review*, 14(3).

Hoch, S., & Mügge, D. (2026). “EU Influence in Global AI Governance and its Limits.” *Digital Society*, 5, Article 45.

European Union. Regulation (EU) 2024/1689, Artificial Intelligence Act, Article 2.