How to navigate the implementation of the AI Act, which came into force at the start of August. (Photo: Shutterstock)

How to navigate the implementation of the AI Act, which came into force at the start of August. (Photo: Shutterstock)

Does a company or a non-profit organisation have to disclose that it uses artificial intelligence (AI)? In most cases, no. The AI Act does not require a general declaration. However, it does oblige certain parties to inform affected individuals, to register certain high-risk systems or to report an incident. Here’s the lowdown in 13 questions… and 13 answers.

Using artificial intelligence (AI) to correct an email does not trigger the same obligations as deploying a chatbot to interact with customers or using an algorithm to sort through job applications. The European regulation does not require a “we use AI” label to be displayed everywhere. It targets specific uses, specific individuals and, in some cases, specific authorities.

1. Is a company or a non-profit organisation required to declare all its AI tools to an authority?

No. There is no general register in which every business or organisation is required to register ChatGPT, Copilot, translation software or a management tool featuring an AI function. An organisation that uses these tools for routine tasks is therefore not, on that basis alone, required to submit a form to the CNPD or any other authority.

It must, however, be familiar with its uses, ensure that no prohibited practices are being carried out, and provide adequate training to those who use AI on its behalf. This inventory is an internal control measure; it is not a public statement.

2. What exactly is a “high-risk system”?

It is not a particularly powerful AI system, nor is it a tool that is likely to make many mistakes. It is a system used in one of the sensitive situations specifically targeted by the AI Act, because it can have a significant impact on a person’s safety, rights or opportunities.

For a company or a non-profit organisation, the most telling examples are software that sorts or scores job applications, assesses employees, decides on promotions or dismissals, determines eligibility for training, assesses eligibility for financial assistance or essential public services, establishes an individual’s creditworthiness, or evaluates risk and premiums for life or health insurance. Certain applications of biometrics, critical infrastructure, the justice system, policing, migration and elections also fall into this category. A second category covers AI integrated as a safety component into certain regulated products that must undergo a conformity assessment, for example certain medical devices or machines.

The purpose matters more than the name of the software. A tool that sorts CVs alphabetically is not necessarily high-risk; the same tool, if it ranks candidates according to their suitability, may be. Even where a particular use is included in the list, it may not fall within that category if it does not materially influence the decision and does not pose a significant risk. This exception does not apply to a system that profiles an individual.

3. Who is required to register an AI system in an official database?

Registration applies to certain high-risk systems, not to all devices. The supplier of a high-risk system listed in Annex III must, in principle, register their organisation and system in the European database before placing it on the market or putting it into service. A public authority, a European institution or a person acting on its behalf must also register its use before deploying such a system.

A private company or a non-profit organisation that purchases a high-risk system is therefore not automatically responsible for registering it. It must, however, check that the product is registered where required. An organisation that develops the system, markets it under its own name or changes its intended use may become a supplier and be subject to additional obligations.

Following the entry into force of the AI Omnibus, these rules have been deferred until 2 December 2027. For AI integrated into certain regulated products, the deadline is 2 August 2028.

4. Should we tell customers or members that a chatbot is an AI?

Yes, when people interact directly with an AI system and this is not immediately obvious. The information must be clear, accessible and provided no later than the first interaction. A visible notice such as “You are interacting with an automated assistant using artificial intelligence” better serves this purpose than a clause buried in the terms and conditions.

The obligation to design the system rests with the supplier. However, the company or non-profit organisation installing the chatbot must ensure that the interface displays the message. If it displays the message under its own name, its legal role must be reassessed.

5. Does a company have to notify all its customers? Does a non-profit organisation have to notify all its members?

No. The AI Act does not require a general notification to be sent to all customers or members. It is based on exposure and usage, not on the commercial relationship or membership status. The individual concerned must be informed if they are interacting with a chatbot that is not obviously automated, if they are exposed to an emotion recognition or biometric categorisation system, or, ultimately, if a decision concerning them is made or assisted by certain high-risk systems.

A company is therefore not required to write to all its customers to inform them that its marketing department is using AI to draft a slogan. Similarly, a non-profit organisation does not need to notify all its members that its secretariat is using AI to reword an invitation or translate an announcement. However, the GDPR may require such notification if personal data is being processed.

6. Should it be stated that an image, video or audio clip has been generated by AI?

Yes, when a company or organisation publishes a “hyper-fake” – that is, content that has been generated or manipulated to resemble a real person, object, place or event and could be mistaken for the real thing. The publisher must state that the content has been artificially generated or manipulated. In the case of a work that is clearly artistic, creative, satirical or fictional, the information may be adapted so as not to detract from its presentation.

For its part, the provider must ensure that summarised content is machine-readable. This technical tagging and the visible information provided by the publishing organisation are two separate obligations.

7. Should a text produced using AI carry a disclaimer?

Not in all cases. The requirement applies to text generated or manipulated by AI that is published with the aim of informing the public on a matter of public interest. Even in such cases, the regulation provides for an exception where the content has been subject to human review or editorial oversight and a natural or legal person assumes editorial responsibility for the publication.

A draft that has been proofread, checked and endorsed by the organisation does not, therefore, automatically warrant an “AI-generated” label. Copying and pasting, without checking, a text relating to current affairs, public health, local politics or any other subject of public interest increases the organisation’s obligation to be transparent – and, above all, poses an editorial risk.

8. Should people who are being filmed or analysed be informed?

Yes, when an emotion recognition or biometric categorisation system is used in a permitted context. Data subjects must be informed of how the system works. Certain biometric or emotion recognition practices are, moreover, prohibited or subject to strict regulations; therefore, providing information alone is not sufficient to render their use lawful. The GDPR also continues to apply to personal and biometric data.

9. Is a company required to notify its employees?

In the case of a high-risk system used in the workplace, the employer must inform the workers’ representatives and the employees concerned before it is put into use. This rule applies in particular to certain recruitment, task allocation or assessment systems.

According to the current timetable, the specific obligations relating to these systems will only apply from 2 December 2027. However, this does not permit a lack of transparency in their roll-out until then: employment law, the GDPR and national rules on information and consultation may already require certain steps to be taken.

10. Should a client, candidate or beneficiary be informed when an AI system is involved in a decision?

Where a high-risk system makes or assists in making a decision concerning an individual, the operator must inform that individual. This may relate to recruitment, education or certain essential services. This obligation will come into force on 2 December 2027.

This information does not replace either the right to receive an explanation regarding certain decisions or the obligations under the GDPR.

11. Who should report an incident?

In the case of a high-risk system, the deployer who identifies a serious incident must immediately inform the supplier, followed by the importer or distributor and the competent authority. If normal use presents a risk, they must alert the same parties and suspend use. The supplier bears the primary responsibility for formal notification; if the operator is unable to contact the supplier, the procedure applies to the operator by analogy.

It is not a question of reporting every single error. However, the organisation must be able to recognise an incident and alert the relevant people.

12. Who should you contact in Luxembourg?

As at 6 August 2026, Bill 8476, which is intended to designate the competent Luxembourg authorities, is still “before committee”. It provides for a central role for the CNPD and powers for several sectoral supervisors.

13. What should be done now?

The first step is not a general statement. You need to draw up a list of the tools used, noting who supplies them, what they do, which individuals they affect and what data is entrusted to them. For each use case, the organisation must then decide on four issues: should the individual be informed, should the system be logged, should the use be documented for internal purposes only, or should an incident be reported?

Companies and non-profit organisations are not required to report everything. They must provide the right information to the right person at the right time.