The logic of the Brussels Effect is straightforward. The EU is a market too large and too rich to ignore. When it sets strict rules for products sold within its borders, multinational companies often find it cheaper to apply the EU standard everywhere than to maintain two versions of a product. The EU rule thus becomes the global default — not because anyone signed a treaty, but because the economics of a single global product line make compliance contagious. This is regulatory power projected through market size rather than diplomacy.
How the effect actually spreads
Bradford distinguishes two channels. The de facto Brussels Effect occurs when companies voluntarily standardise on the EU rule worldwide because splitting their product is impractical — as many did with data-handling practices after the GDPR. The de jure effect follows when other governments, seeing that their firms already comply with EU rules, simply copy those rules into their own law, both to help their exporters and because the EU has done the drafting work for them. Privacy law spread across dozens of countries this way.
The EU AI Act as the test case
The EU AI Act, adopted in 2024, is the most ambitious attempt yet to trigger the effect for artificial intelligence. It regulates AI by risk tier: banning certain 'unacceptable-risk' uses outright, imposing strict obligations on 'high-risk' systems, and adding specific transparency and systemic-risk rules for the most capable general-purpose models. Because it applies to any provider placing a system on the EU market, its reach extends well beyond Europe's borders — a US or Chinese developer selling into the EU must comply.
If the Brussels Effect holds, the Act's requirements — risk assessment, documentation, transparency, and obligations for the most capable models — could become the baseline that global developers build to by default, dragging worldwide practice upward without a single line of treaty text. That is a genuinely significant lever, and it is already shaping how companies document and govern their systems.
Why it is not enough for existential risk
- Market access is the wrong lever for the worst risks. The Brussels Effect works by conditioning sales into the EU. But a lab racing to build superintelligence for strategic advantage is not primarily motivated by selling a product into the European market — and a system developed and deployed outside the EU need not comply at all.
- It cannot reach a state actor. Regulation of market access does nothing to constrain a government-backed program that never sells to European consumers. The most dangerous development could route entirely around it.
- It is unilateral, not verified. The EU sets the rule, but there is no international inspection regime confirming that frontier developers worldwide actually meet it — only the pressure of market access.
- It can provoke fragmentation. Rival powers may reject the EU standard rather than adopt it, especially where they see it as a competitive or geopolitical imposition, splitting the world into regulatory blocs instead of converging it.
The Brussels Effect is one of the most powerful tools we have for raising the global floor on AI. But it works by controlling who can sell into Europe — and the actor racing to build something too dangerous to sell is exactly the one it cannot touch.
Naoto Nakada, Founder · Nakada Foundation to Save Humanity
A lever, not the whole machine
The Brussels Effect deserves to be used to its fullest. It can standardise safety documentation, evaluation practices, and transparency across the commercial AI industry faster than any treaty could, and it builds the regulatory habits that a binding regime would later formalise. But it is a mechanism for governing AI as a marketed product, not for governing the frontier race as a strategic contest. The catastrophic scenarios — a state or lab crossing a capability line for advantage, outside any market Europe controls — sit precisely in its blind spot. The lesson is to deploy the Brussels Effect aggressively for what it can do, while recognising that binding international agreement, with verification and enforcement, is required for what it cannot. Market power raises the floor; only a treaty can hold the ceiling.
Common questions.
A term coined by scholar Anu Bradford for the way the European Union sets de facto global standards by regulating access to its large single market. When the EU imposes strict rules, multinational companies often apply the EU standard worldwide because maintaining two product versions is impractical, and other governments frequently copy EU rules into their own law. It is regulatory power projected through market size rather than treaties.
Partly. The EU AI Act applies to any provider placing an AI system on the EU market, so its risk-tier obligations — assessment, documentation, transparency, and rules for the most capable models — could become the global baseline that developers build to by default. This is a powerful lever for raising standards across the commercial AI industry without a treaty, and it is already influencing how companies govern their systems.
Because it works by conditioning access to the European market, and the most dangerous development — a lab or state racing to build superintelligence for strategic advantage — is not primarily motivated by selling into the EU and can route around the rule entirely. It cannot reach a government-backed program that never sells to European consumers, it involves no international verification, and it may fragment the world into rival regulatory blocs rather than converging it.
The European Union's 2024 law regulating AI by risk tier: it bans certain unacceptable-risk uses, imposes strict obligations on high-risk systems, and adds transparency and systemic-risk requirements for the most capable general-purpose models. It applies to any provider placing a system on the EU market, giving it reach beyond Europe — the primary vehicle through which the EU hopes to trigger a Brussels Effect for AI.