EU AI Act for US companies

Checked against the AI Act as amended by the Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since July 27, 2026).

The EU AI Act can apply to a US company with no EU office. Placing an AI system on the EU market, or its output being used there, can be enough (Art. 2(1)). Duties depend on your role and the risk tier, system by system. Some rules already apply; most high-risk duties start December 2, 2027 or August 2, 2028. The dates come from Art. 113 and Art. 111.

From	What applies	Who it binds
February 2, 2025	Prohibited practices (Art. 5) and AI literacy (Art. 4)	Bans: everyone in scope. Literacy: providers and deployers
August 2, 2025	General-purpose AI model duties (Arts. 53 to 55)	Model providers
August 2, 2026	Transparency (Art. 50) and most other rules	Providers and deployers of the systems Art. 50 lists
December 2, 2026	Two new bans (Art. 5(1)(ba)-(bb)); output marking for generative systems already on the market (Art. 111(4))	Everyone in scope; providers of those generative systems
August 2, 2027	Model duties for models placed on the market before August 2, 2025 (Art. 111(3))	Model providers
December 2, 2027	High-risk duties for Annex III systems	Providers, deployers, importers, distributors, authorized representatives
August 2, 2028	High-risk duties for Annex I systems	The same

Apart from Art. 5, the Act reaches a high-risk system placed on the market before its date only after a significant design change, except systems for public authorities, due by August 2, 2030 (Art. 111(2)).

When a US company is in scope

The Act (Regulation (EU) 2024/1689) covers “AI systems”: machine-based systems that, with some autonomy, infer from their input how to generate outputs such as predictions, content, recommendations or decisions (Art. 3(1)). Software that only runs rules people wrote may fall outside.

Under Art. 2(1), a US company is in scope when:

it places an AI system or a general-purpose AI model on the EU market, or puts a system into service there, including for its own use, wherever it is based (2(1)(a), 3(11));
it, or its EU subsidiary or branch, uses AI as a deployer established or located in the EU (2(1)(b));
it is a provider or deployer outside the EU and the system’s output is used in the EU (2(1)(c)); or
its EU entity imports AI systems, it distributes AI systems on the EU market, or it places on the EU market a product with an AI system built in under its own name (2(1)(d)-(e)).

Exclusions include uses only for military, defense or national security (2(3)), systems built only for scientific research (2(6)), testing before market other than real-world testing (2(8)), and open-source systems that are not high-risk or under Art. 5 or 50 (2(12)). Each route, with examples: Does the EU AI Act apply to a US company?

Your role, system by system

Roles are set per system, and one company can hold several (Art. 3(3)-(7)):

Provider: develops a system, or has it developed, and places it on the market or puts it into service under its own name or trademark.
Deployer: uses a system under its authority, outside personal, non-professional use.
Importer: based in the EU, places on the EU market a system carrying a non-EU company’s name.
Distributor: anyone else in the supply chain, other than the provider or importer, who makes a system available on the EU market.
Authorized representative: acts in the EU under a provider’s written mandate. Non-EU providers need one for high-risk systems, from the high-risk dates above (Art. 22), and for general-purpose AI models, on the model dates above, except some open-source models (Art. 54).

Art. 25 shifts. From December 2, 2027 (Annex III) or August 2, 2028 (Annex I), you become the provider of a high-risk system if you put your name or trademark on it (without prejudice to contracts that allocate the obligations otherwise), substantially modify it so it stays high-risk, or change the intended purpose of a system that was not high-risk so that it becomes high-risk (Art. 25(1)). Common setups: Provider or deployer: your EU AI Act role.

The risk tiers

Systems that are not banned, not high-risk and outside Art. 50 (often called minimal risk) carry no tier-specific duties. AI literacy applies to all of them.

Prohibited practices (Art. 5)

Banned since February 2, 2025, each under the conditions in its text (Art. 5): manipulation or exploiting vulnerabilities that causes, or is reasonably likely to cause, significant harm; certain social scoring; crime prediction based solely on profiling or personality traits; untargeted facial image scraping; emotion recognition at work or school, except for medical or safety reasons; biometric categorization to infer sensitive traits; and real-time remote biometric identification in public spaces for law enforcement, with narrow exceptions. From December 2, 2026, two more bans cover realistic intimate imagery of identifiable people made without their explicit consent, and child sexual abuse material (5(1)(ba)-(bb), limited by 5(1a)-(1b)).

High-risk systems (Art. 6)

There are two routes to a high-risk AI system (Art. 6):

Annex III. Uses listed in Annex III, read point by point: biometrics, critical infrastructure, education, employment, essential services (such as credit scoring of individuals, except fraud detection), law enforcement, migration, and justice and democratic processes. Tools that filter job applications or evaluate candidates are listed (point 4(a)).
Annex I. The system is a safety component of a product, or is itself a product, under the EU product laws in Annex I, such as medical devices, and the product needs third-party conformity assessment for health and safety reasons (6(1), 6(1c)). A system used only for non-safety aspects such as user assistance, efficiency, automation, convenience or quality control is not a safety component, unless its failure would endanger health and safety (6(1a)-(1b)). For Section B products (vehicles, aviation, rail, marine, machinery), only a few AI Act articles apply directly (Art. 2(2)); the AI requirements reach them through those sector laws.

An Annex III system is not high-risk if it poses no significant risk of harm and meets one of four conditions in Art. 6(3), such as a narrow procedural task. This never covers systems that profile people, and the provider must document that assessment and register the system (6(4)). It cannot lift an Annex I classification.

From the high-risk dates, providers carry most duties (Art. 16): risk management, data governance, documentation, logging, human oversight, quality management, conformity assessment and more. Deployers use the system as instructed, assign competent human oversight, check that input data they control is relevant, monitor the system and report risks and serious incidents (suspending use where it presents a risk), keep the logs they control for at least six months, inform workers and their representatives before workplace use, and, for Annex III decisions about people, inform those people, among other duties (Art. 26). For Annex III systems other than critical infrastructure, public bodies, private entities providing public services, and deployers of credit scoring or life and health insurance pricing also assess the impact on fundamental rights before first use (Art. 27).

Transparency (Art. 50)

Since August 2, 2026 (Art. 50):

Providers design systems that interact directly with people so people know it is AI, unless obvious (50(1)).
Providers of generative systems mark output as AI-generated in a machine-readable way, except for assistive editing or output that does not substantially alter the input (50(2)); systems already on the market have until December 2, 2026 (Art. 111(4)).
Deployers of emotion recognition or biometric categorization inform the people exposed (50(3)).
Deployers disclose deep fakes (for evidently artistic, satirical or fictional work, a lighter disclosure), and disclose AI-generated text published to inform the public on matters of public interest unless a person reviewed the text and someone holds editorial responsibility (50(4)).

Narrow exceptions cover uses authorized by law to fight crime.

General-purpose AI models

Since August 2, 2025, model providers keep documentation, inform downstream builders, keep a copyright policy and publish a training-content summary (Art. 53), with more duties for systemic-risk models (Art. 51, 55). Models already on the market by then have until August 2, 2027. Building on another company’s model through its API usually makes you the provider or deployer of your own system; the model duties stay with the model’s provider. If you fine-tune or otherwise modify the model itself, you can become the provider of the modified model, with duties limited to your changes (Recital 109); the Commission’s guidelines on general-purpose AI models explain when.

AI literacy (Art. 4)

Every provider and deployer in scope must take measures to support the AI literacy of its staff and of other people who operate or use AI systems on its behalf, whatever the system’s risk tier (Art. 4). Since the Omnibus, the text says this does not require them “to guarantee any specific level of AI literacy of any individual”.

Penalties

Member states set penalties (Art. 99). Maximum fines, whichever figure is higher: EUR 35 million or 7% of worldwide annual turnover for prohibited practices (99(3)); EUR 15 million or 3% for the duties listed in 99(4), including those of high-risk providers and deployers, importers, distributors and Art. 50. SMEs and start-ups face the lower figure (99(6)). From August 2, 2026, the Commission can fine model providers up to EUR 15 million or 3%, whichever is higher (Art. 101).

What to do first

List your AI systems, including AI features in vendor tools, and note how each reaches the EU.
Set your role per system and check the Art. 25 shifts.
Rule out the Art. 5 practices and check Art. 50 for live chatbots and generated content.
Put AI literacy measures in place.
Classify each system against Annex III and Annex I, and record its market date. The Commission’s high-risk classification guidelines are still a draft.

How Oblioria helps

Your team answers a structured questionnaire for each AI system, and a consultant assigned to your company reviews the answers and can ask follow-up questions. Your report then sets out, from your answers and after expert review, your likely role, the risk category and the obligations that apply, with their dates. It is general information, not legal advice, and Oblioria is not a law firm. See how it works.

FAQ

Does the AI Act apply if we have no office in the EU? It can: placing a system on the EU market, or its output being used there, is enough (Art. 2(1)(a), (c)). See Does the EU AI Act apply to a US company?

We only use AI tools others built. What is our role? Usually deployer, unless an Art. 25 shift applies. See Provider or deployer: your EU AI Act role

Is AI in hiring high-risk? Recruitment, promotion, termination and performance tools are listed in Annex III point 4, from December 2, 2027. The Art. 6(3) exception is closed if the tool profiles people. New York City has its own rules for hiring tools: NYC Local Law 144.

Does the AI Act change our GDPR duties? It leaves the GDPR in place: apart from Arts. 4a and 59 (some processing for bias detection and in regulatory sandboxes), the AI Act does not affect it (Art. 2(7)). Both laws can apply to the same system. See GDPR for US companies with EU users.

Sources

AI Act Service Desk, text as amended by the Digital Omnibus: Art. 2, Art. 3, Art. 4, Art. 5, Art. 6, Art. 16, Art. 22, Art. 25, Art. 26, Art. 27, Art. 50, Art. 51, Art. 53, Art. 54, Art. 55, Art. 99, Art. 101, Art. 111, Art. 113, Annex I, Annex III, Recital 109
AI Act Service Desk, Guidelines on the classification of high-risk AI systems (draft)
European Commission, Guidelines on the scope of obligations for providers of general-purpose AI models
Regulation (EU) 2024/1689, Official Journal
Regulation (EU) 2026/1744 (Digital Omnibus on AI), Official Journal, published July 24, 2026

This page is general information, not legal advice.

The EU AI Act for US companies