What is extraterritoriality in AI law?

How it works

<strong>Scope begins with a jurisdictional hook</strong>

In AI regulation, extraterritoriality rarely depends on nationality alone. It normally depends on a hook to the market or to protected persons. Typical hooks are placing an AI system or general-purpose AI model on the market, putting it into service there, doing business there, or having the system's result used there. The EU AI Act is the clearest current model. It applies to providers that place AI systems or general-purpose AI models on the Union market whether they are established in the Union or in a third country. It also reaches providers and deployers in a third country where the result produced by the AI system is used in the Union.

That design matters because it moves the scoping question away from server location. A system can be trained and run abroad yet still fall inside a rulebook because the legally relevant link is the market, the deployer, the distributor or the use of the result inside the jurisdiction.

<strong>Role mapping comes before compliance</strong>

Once the hook is clear, the next step is role mapping. Under the EU AI Act, the provider is the party that develops, or has developed, an AI system or general-purpose AI model and places it on the market or puts it into service under its own name or trade mark. The deployer is the party that uses the system under its authority. The distributor is a supply-chain actor, other than the provider or importer, that makes the system available on the market.

This matters because geography and role are separate questions. A company outside the EU may still be a provider for EU purposes. A local reseller may think of itself as a distributor, but if it changes the legal posture of the product it may trigger a different status. Good governance therefore needs a jurisdiction-by-jurisdiction role map rather than one internal label used everywhere.

<strong>Supply-chain design makes extraterritoriality real</strong>

Extraterritoriality becomes practical through supply-chain controls. The EU AI Act puts importers and distributors directly into scope. For non-EU providers of high-risk AI systems and non-EU providers of general-purpose AI models, it also requires an authorised representative established in the Union before the relevant product is placed on the market. The Commission's guidance for general-purpose AI makes the market-placement point even clearer by treating availability through APIs, downloads, cloud services, integration into applications, and certain internal processes for services offered to third parties as ways a model may be placed on the Union market.

This is why cross-border AI compliance often arrives through sales channels before it arrives through a regulator letter. An organisation may not "enter" a market by opening an office. It may enter by serving customers through an API, appointing a distributor, embedding a model in a partner's application, or allowing a local customer to rely on the system in a regulated workflow.

<strong>Modification and rebranding can move liability</strong>

Legal role can also change after launch. Under the EU AI Act, a distributor, importer, deployer or other third party can become the provider of a high-risk AI system if it puts its own name or trade mark on the system, makes a substantial modification, or changes the intended purpose so that the system becomes high risk. In other words, regulatory responsibility can move with the business model.

This is one reason extraterritoriality is often misread in practice. Teams focus on who trained the original model, but the law may focus on who is now marketing it, modifying it, or using it in a way that changes its regulated status. Contracts remain essential for sharing documents, technical support, incident notice and change control, but they do not erase the statutory role a regulator assigns.

<strong>Enforcement depends on local institutions and records</strong>

A broad scope clause only works if there are institutions and records to support it. In the EU, supervision is built around national authorities, market surveillance, conformity procedures, document retention and the Commission's oversight of general-purpose AI obligations. In Colorado, the current state AI law ties duties to persons doing business in the state and gives the attorney general exclusive enforcement authority. In China, current public-facing generative AI rules attach to services provided to the public within mainland China and, for certain services, to security assessment and filing.

For operators, this means scope analysis should start before launch, not after launch. Once the system is sold into a market, integrated by a local partner, or used for decisions affecting people in that territory, the practical enforcement path may already exist through local authorities, local representatives, local distributors or local customers.

<strong>Standards and treaties shape the practical picture</strong>

Standards do not decide whether a statute applies, but they do create the evidence needed to answer extraterritorial questions. NIST's AI RMF expects organisations to define key terms and intended uses, align governance to legal standards, document legal review processes, include third-party AI systems in governance, and build monitoring and impact-assessment practices. That kind of record helps answer the questions regulators and large buyers actually ask: who is the provider, what is the intended purpose, which jurisdictions matter, what changed after fine-tuning, and who can produce the evidence.

International instruments play a different role again. The Council of Europe Framework Convention on AI works through state obligations. It asks participating states to implement principles, safeguards, remedies, and risk and impact assessment duties through domestic law. That can drive convergence across countries, but it is not the same as a domestic statute that directly reaches a foreign provider, deployer or distributor because it accesses the market.