Jorge Carvalho da Silva

Authorised Representative for Artificial Intelligence in the European Union

Representing providers of artificial intelligence systems and models established outside the European Union, under the AI Act.

The EU Artificial Intelligence Act (EU AI Act) requires providers established outside the European Union, in certain cases, to appoint an authorised representative within the Union. The representative acts as the point of contact with the European authorities and carries out the document-verification and regulatory-cooperation tasks that the law assigns to the role.

I carry out this role drawing on a career that combines legal practice, founding and running companies, and specialist training in Artificial Intelligence.

The responsibilities of an authorised representative call for more than a presence in the European Union. They require knowledge of the legal framework, an understanding of how businesses work and the ability to perform, independently, the tasks the Regulation sets out.

Who needs an authorised representative in the European Union?

Whether an appointment is required depends on the nature of the artificial intelligence system or model and on the legal regime that applies to it.

Providers of high-risk AI systems

Article 22 of the Regulation requires certain providers of high-risk artificial intelligence systems established outside the European Union to appoint an authorised representative.

Whether a system counts as high-risk depends on the criteria in Article 6 and Annexes I and III of the Regulation.

The analysis should therefore start by identifying the system, its intended purpose and the regulatory framework that applies to it.

Providers of general-purpose AI models — GPAI

Article 54 sets out a specific representation requirement for providers of general-purpose artificial intelligence models established outside the European Union.

These models, which are often the foundation for a range of applications and systems, are subject to their own regime of documentation, transparency and regulatory cooperation.

The obligation to appoint an authorised representative is subject to exceptions, notably for certain models released under a free and open-source licence, on the conditions set out in Article 54(6).

The exception does not apply to general-purpose AI models with systemic risk.

Whether an authorised representative must be appointed has to be determined case by case. Not every non-EU operator of Artificial Intelligence is subject to the obligation.

What does an authorised representative do?

The authorised representative performs the tasks set out in the Regulation, as specified in the written mandate.

The content of those tasks depends on whether the representation concerns high-risk AI systems or general-purpose AI models.

Document verification

The authorised representative checks, as the applicable rules require, that the provider has produced the technical documentation and the other items required of it.

For high-risk systems this includes, in particular, the EU declaration of conformity, the technical documentation and the conformity assessment procedures.

For GPAI models it covers the technical documentation and verification that the relevant regulatory obligations have been met, including those in Article 53 and, where applicable, Article 55.

Keeping and making available the documentation

The authorised representative keeps the documents required by the Regulation for the periods the law prescribes and ensures that they can be produced to the competent authorities.

The documentation must be organised so that it is possible to identify the systems or models covered by the mandate, the relevant versions and everything needed to meet the regulatory obligations.

Liaison with the European authorities

The authorised representative is the formal point of contact for the competent authorities.

It receives and follows up reasoned requests for information and documentation, coordinating with the provider and responding within the scope of the responsibilities entrusted to it.

Regulatory cooperation

The authorised representative cooperates with the authorities on actions concerning the systems or models covered by the mandate, as provided for in the Regulation.

This requires clear lines of communication and the availability to see the necessary procedures through.

Terminating the mandate

The Regulation provides for situations in which the authorised representative must terminate the mandate, in particular where it has reason to believe that the provider is acting in breach of its regulatory obligations.

In that event it must also inform the competent authority, as the law requires.

This duty shows that the authorised representative is not simply the provider’s administrative agent: it performs a role with responsibilities of its own under the European regulatory framework.

What it does not do

It does not prepare the technical documentation or the declaration of conformity. It does not carry out the conformity assessment. It does not assess technical merit.

It does not guarantee compliance: for systems, it checks that the required documents exist and appear complete; for models, that the obligations have been met.

It does not represent providers engaged in practices prohibited by Article 5 of Regulation (EU) 2024/1689.

How is the representation put in place?

  1. Initial contact and identifying the provider

    The process starts with identifying the company, the country in which it is established and the artificial intelligence systems or models it intends to place on the European market.

  2. Analysing the regulatory framework

    We analyse the nature of the system or model, the legal regime that applies and whether an authorised representative must be appointed.

    At this stage we also identify the main documents and information needed for the representation.

  3. Proposal and written mandate

    Once the scope of the representation is settled, the terms of the engagement are agreed and the written mandate is put in place.

    The mandate identifies the systems or models covered, the tasks undertaken, the parties’ responsibilities and the communication procedures.

  4. Starting and managing the representation

    Once the mandate is in place, the relevant documents are organised and the lines of contact between the provider, the authorised representative and the competent authorities are established.

    Follow-up continues for the duration of the mandate, in line with the regulatory obligations and the agreed terms.

Responsibilities of the provider and of the authorised representative

Appointing an authorised representative does not replace the obligations that the Regulation places on the provider.

The provider remains responsible for meeting the requirements that apply to it.

The authorised representative, for its part, is answerable for the tasks the Regulation assigns to it and that are specified in the mandate.

This distinction is essential to defining the scope of the representation and each party’s responsibilities correctly.

The representation is neither a certificate of compliance nor a guarantee that no regulatory risk exists.

My role

My career combines legal experience with founding, developing and managing companies.

To that experience I have added specific training in Artificial Intelligence, including a postgraduate programme in Communication and Artificial Intelligence and advanced training in Artificial Intelligence Law and Practice.

My work in this area centres on the European legal framework for Artificial Intelligence, on the obligations of the different operators, and on the questions of responsibility and governance raised by the use of increasingly autonomous systems.

The Regulation does not reserve the role of authorised representative to any profession.

However, the responsibilities attached to the role call for a rigorous understanding of the legal obligations, the ability to review documentation and independence in assessing situations that could jeopardise compliance with the Regulation.

It is by combining legal experience, business knowledge and training in Artificial Intelligence that I approach my work as an authorised representative in the European Union.

Frequently asked questions

Does my company need to appoint an authorised representative?

It depends on the nature of the artificial intelligence systems or models it makes available and on the applicable legal regime. The obligation does not apply across the board to every provider established outside the European Union.

Can an authorised representative established in Portugal act for a company before the authorities of other Member States?

The Regulation provides for the appointment of an authorised representative located or established in the European Union. The precise scope of its tasks before the authorities depends on the applicable regime and on the terms of the mandate.

What is the difference between representation for high-risk systems and for GPAI models?

They are separate regimes, set out in Articles 22 and 54 of the Regulation, with different conditions, documentation requirements and competent authorities.

What documentation has to be provided?

It depends on the system or model and on its regulatory framework. Identifying it is one of the first steps of the analysis.

Does the authorised representative take on the provider’s responsibilities?

No. It does not take the provider’s place in its compliance obligations. It does, however, bear responsibilities of its own for carrying out the tasks the law assigns to the authorised representative.

What happens when a system or model is updated?

Relevant changes must be notified and assessed to determine their effect on the documentation, the regulatory obligations and the scope of the mandate.

Does an open-source model need an authorised representative?

No, unless it presents systemic risk (Article 54(6) of Regulation (EU) 2024/1689). The exception has cumulative conditions and requires that the model is not monetised (recital 103 of Regulation (EU) 2024/1689). For high-risk systems, being open source does not remove the obligation.

My high-risk system is already on the market. What changes in 2027?

As a rule, an appointment does not become mandatory: under Article 111(2) of Regulation (EU) 2024/1689, the Regulation applies to systems placed on the market before the relevant dates of application only if they are subsequently subject to significant changes in design. There are three caveats: practices prohibited by Article 5 never benefit from this transitional relief; systems intended for use by public authorities must comply by 2 August 2030; and you must be able to prove the date of first placement on the market, by type and model (recital (39) of Regulation (EU) 2026/1744).

Is the authorised representative’s name made public?

As a rule, yes. For high-risk systems it appears in the EU declaration of conformity (Annex V, point 2, of Regulation (EU) 2024/1689) and, for models, in the public summary of training content. This is one reason why the choice of authorised representative deserves careful thought.

Model or system: what is the difference?
The model is the engine

A general-purpose AI model is trained on a large volume of data and competently performs a wide range of different tasks, and it can be integrated into a great variety of systems and applications (Article 3, point 63, of Regulation (EU) 2024/1689). There are language, image, audio and code models. It is what powers the assistants everyone knows. The company that develops the model and places it on the market is the model provider, and its authorised representative is the one under Article 54 of Regulation (EU) 2024/1689.

The system is the product

An AI system has a specific purpose and is used by whoever deploys it. A conversational assistant such as ChatGPT or Gemini is a system built on a model. The same model can underpin many systems from different providers.

When a system is high-risk

When it is a safety component of a regulated product (Annex I, such as medical devices) or is intended for one of the areas listed in Annex III (Article 6 of Regulation (EU) 2024/1689). Examples by area:

  • Biometrics: remote biometric identification of individuals.
  • Critical infrastructure: management of road traffic and of the supply of water, gas, heating and electricity.
  • Education and vocational training: admission to educational institutions, assessment of learning outcomes, monitoring of exams.
  • Employment: recruitment and screening of applications, decisions on promotion or termination, performance monitoring.
  • Essential services: credit scoring of individuals, risk assessment and pricing in life and health insurance, eligibility for public benefits, triage of emergency calls.
  • Law enforcement: assessment of risks and of evidence by authorities.
  • Migration, asylum and border control: risk assessment and examination of visa and asylum applications.
  • Administration of justice and democratic processes: support for judicial authorities in interpreting facts and the law.
Classification is a case-by-case matter

It depends on the specific purpose of the tool, which is what the preliminary analysis is for. Whether an authorised representative is needed, and under which provision, depends on whether the point concerns the model (Article 54 of Regulation (EU) 2024/1689) or the system (Article 22 of Regulation (EU) 2024/1689).

Contact

Representing, in the European Union, providers of Artificial Intelligence established outside the Union.

For professional enquiries, requests for information, collaboration proposals or the appointment of an authorised representative in the European Union, please write to me at the email address below, briefly stating the subject and what you are seeking.

Contact →