Guide

Do I need an AI Act representative?

You need an AI Act authorised representative if your company is established outside the EU and either places a general-purpose AI model on the EU market (Article 54, in force since 2 August 2025) or makes a high-risk AI system available there (Article 22, from 2 December 2027 for Annex III systems). If you only build features on another provider's model, without changing the model so much that you become its provider, and your system is not high-risk, you do not need one.

By Julian Gage Last reviewed: 31 October 2026

The rule

The EU AI Act has two representative duties, and they catch different companies.

Article 54, general-purpose AI models. "Prior to placing a general-purpose AI model on the Union market, providers established in third countries shall, by written mandate, appoint an authorised representative which is established in the Union." Chapter V, which holds Article 54, has applied since 2 August 2025, so this duty is live now.
Article 22, high-risk AI systems. Providers established in third countries have to appoint an authorised representative established in the Union, by written mandate, before they make a high-risk AI system available on the Union market. After the Digital Omnibus on AI (Regulation (EU) 2026/1744), this applies from 2 December 2027 for high-risk systems under Annex III and from 2 August 2028 for high-risk systems built into products under Annex I.

In both cases the representative is established in the EU and holds your documentation for the authorities for 10 years after the model or system is placed on the market. No particular Member State is required.

Who needs one

You need an AI Act authorised representative when all of these are true:

You are the provider. That is the company that develops the model or system, or has it developed, and puts it on the market under its own name.
You are not established in the EU. If your only European office is in Norway, Iceland or Liechtenstein, you still count as not established in the EU. Those countries are in the EEA but not the EU, and the AI Act only counts an establishment in the EU.
You place a general-purpose AI model on the EU market (Article 54), or make a high-risk AI system available there (Article 22). A system is high-risk when it is a safety component of a product covered by the Annex I legislation under Article 6(1), or falls in one of the Annex III use cases under Article 6(2).

If you do both, you need both appointments. They are two separate duties.

The Article 22 duty can also move to you. Under Article 25(1), a company that puts its own name or trademark on someone else's high-risk system, makes a substantial modification that keeps it high-risk, or changes a system's intended purpose so that it becomes high-risk is treated as its provider.

Who does not need one

Companies that build on another provider's model. Running a feature on a third-party model makes you a deployer, or at most a downstream provider of a system built on that model. The Article 54 duty sits with whoever provides the model.
Providers of open-source general-purpose AI models. Article 54(6) exempts a model released under a free and open-source license, with its weights, architecture and usage information made public, unless the model presents systemic risks.
Providers whose system is not high-risk. Article 22 only applies to high-risk systems. Our high-risk classification guide walks through the Annex III and Article 6(1) routes.
Providers established in the EU.

What happens if you do not appoint one

For the high-risk duty, the Act names no fine for the missing appointment on its own. Article 99(4) fines breaches of the provider's duties in Article 16 and of the representative's own duties in Article 22, up to 3 percent of total worldwide annual turnover or a fixed cap, whichever is higher, with a lower cap for SMEs and start-ups under Article 99(6), and a high-risk system made available without the representative Article 22 requires does not meet the Act.

For general-purpose AI models, no article expressly fines a provider for failing to appoint the Article 54 representative. Chapter V is enforced by the Commission against the provider under Article 101, from 2 August 2026, so the duty should not be treated as unenforceable.

What it costs

The Article 54 appointment for general-purpose AI models is €690 a year at the smallest band, for a company with 1 to 10 people, under €2m revenue, and under 5,000 people in the EU and UK whose data you hold. At the three larger bands it is €990, €2,290 and €4,490 a year. That is the price when paid yearly, with no setup fee, and quarterly and monthly payment cost about 5% and 15% more. Whichever of headcount, revenue or the number of people whose data you hold puts you highest sets the band. Every band is on our AI Act representative page.

The Article 22 appointment for high-risk AI systems is €2,200 a year at the smallest band, for a company with 1 to 10 people, under €2m revenue, and under 5,000 people in the EU and UK whose data you hold. At the three larger bands it is €3,157, €7,301 and €14,313 a year. That is the price when paid yearly, with no setup fee, and quarterly and monthly payment cost about 5% and 15% more. Whichever of headcount, revenue or the number of people whose data you hold puts you highest sets the band. Every band is on our high-risk AI authorised representative page.

You can appoint either one online with no sales call.

Check what you need

An AI company is often caught by the GDPR as well, because it processes personal data about people in the EU or the UK. Our free representative checker asks a few short questions and tells you whether you need an EU, UK or Swiss representative under the GDPR, the UK GDPR or the Swiss FADP, and what each would cost. Read it alongside the AI Act test above.

Sources and references

Check which representatives you need

Book a free intro call Start representative sign-up

FAQ

Frequently asked questions

We build on a third-party AI model. Do we need an AI Act representative?

Usually not. Building a feature on another provider's model makes you a deployer, or at most a downstream provider of a system built on that model, and the Article 54 duty sits with whoever provides the model. That changes if you fine-tune or otherwise modify the model so substantially that you become its provider. You would need the Article 22 representative only if your own system is high-risk and you are established outside the EU.

Do we need both the Article 54 and the Article 22 representative?

Only if you both place a general-purpose AI model on the EU market and make a high-risk AI system available there. They are two separate duties, and each needs its own appointment.

When does the high-risk AI representative duty start?

From 2 December 2027 for high-risk systems under Annex III, and from 2 August 2028 for high-risk systems built into products under Annex I, after the Digital Omnibus on AI (Regulation (EU) 2026/1744). The Article 54 duty for general-purpose AI models has applied since 2 August 2025.

Does an office in Norway count as being established in the EU under the AI Act?

No. Norway, Iceland and Liechtenstein are in the EEA but not the EU, and the AI Act only counts an establishment in the EU. A provider whose only European office is in one of them still needs an authorised representative.

Is there an open-source exemption from the AI Act representative duty?

For general-purpose AI models, yes, under Article 54(6), as long as the model is released under a free and open-source license with its weights, architecture and usage information public, and does not present systemic risks. Article 22 has no exemption for high-risk systems.