A municipality buys a chatbot from an American supplier. Its public order department wants event camera footage analysed by a system "also used for national security". An employee drafts a letter to a resident at home with a chatbot. Does the AI Act apply?
The answer starts in Article 2, on scope. The examples are fictional.
In short
The AI Act applies to providers placing AI systems or general-purpose AI models on the EU market or putting AI systems into service in the EU, wherever they are established, to deployers in the EU, and to both roles outside the EU where the output is used here (paragraph 1). Excluded are, among other things, systems used exclusively for military, defence or national security purposes, purely scientific research, pre-market development and the obligations of individuals in purely personal use. In our reading the exclusions are narrow.
Article 2 is the second question. The first is whether there is an AI system (the definition), the third your role; see when you become a provider and the obligations of deployers.
Paragraph 1: seven groups, also outside the EU
The Regulation applies to:
- point (a) providers placing on the market or putting into service AI systems or placing on the market general-purpose AI models in the Union, irrespective of whether they are established in the Union or in a third country;
- point (b) deployers of AI systems that have their place of establishment or are located within the Union;
- point (c) providers and deployers in a third country, "where the output produced by the AI system is used in the Union";
- point (d) importers and distributors;
- point (e) product manufacturers placing on the market or putting into service an AI system with their product under their own name or trademark;
- point (f) authorised representatives of providers which are not established in the Union;
- point (g) affected persons that are located in the Union.
The roles are defined in Article 3, points (3) and (4). A public authority can be a provider; whoever builds a system and uses it under its own name is both.
Points (a) and (c) carry the extraterritorial reach. Recital 22 mentions an EU operator outsourcing work to a party outside the EU that processes data with an AI system and returns the output. That recital says "intended to be used in the Union", point (c) "is used"; in our reading this covers at least output knowingly supplied to an EU recipient.
A provider outside the EU appoints an authorised representative (point (f)) before making a high-risk AI system available on the Union market or placing a general-purpose AI model on it (Articles 22(1) and 54(1)), except for open-source models without systemic risk (Article 54(6)); see our piece on Articles 22 and 54. Affected persons (point (g)) have rights, such as a complaint (Article 85) and an explanation of certain decisions (Article 86).
Paragraphs 2 and 13: regulated products
For high-risk AI systems under Article 6(1) related to products covered by Section B of Annex I, such as vehicles, aviation and since the omnibus machinery, "only Article 6(1), Article 60a and Articles 102 to 112 shall apply"; Articles 57, 58 and 59 only in so far as the requirements have been integrated in the sectoral legislation. See our piece on Articles 102 to 110 and on Article 60a.
The new paragraph 13 concerns Section A of Annex I, such as medical devices: Articles 9 to 15 and 17 to 25 may be limited where the sectoral legislation provides an equivalent or higher level of protection, without reducing the overall level. The Commission must adopt delegated acts for this by 2 August 2027 (paragraph 13, second subparagraph; see also recital 5 of Regulation (EU) 2026/1744); on 22 November 2026 none had appeared in the Official Journal. In our reading no limitation applies until then.
Paragraphs 3 and 4: national security, defence and third countries
The Regulation does not apply outside the scope of Union law and does not affect national security competences of the Member States. It does not apply to AI systems "where and in so far they are placed on the market, put into service, or used with or without modification exclusively" for military, defence or national security purposes, nor to systems not placed on the market or put into service in the Union whose output is used here exclusively for those purposes.
The word "exclusively" is decisive. According to recital 24, a system also used for "for example, civilian or humanitarian purposes, law enforcement or public security purposes" falls within the scope, as does a system with an excluded and a non-excluded purpose.
Paragraph 4 excludes third-country public authorities and international organisations in law enforcement and judicial cooperation with the Union or Member States, provided they offer adequate safeguards for fundamental rights.
Paragraphs 6 and 8: research and development
Paragraph 6 excludes AI systems and AI models, including their output, "specifically developed and put into service for the sole purpose of scientific research and development". Any other AI system usable for research remains covered (recital 25).
Paragraph 8 excludes research, testing or development before placing on the market or putting into service, but not testing in real world conditions; see our piece on Articles 60, 61 and 76. In our reading, for a bank building a model in-house, the AI Act starts at putting into service or the first test outside the lab (Article 3, point (57)); the GDPR applies earlier, also when training on customer data.
Paragraph 10: purely personal use
The Regulation does not apply to "obligations of deployers who are natural persons using AI systems in the course of a purely personal non-professional activity". The provider's obligations remain, such as informing people they are interacting with an AI system (Article 50(1)).
The GDPR does not apply to processing "by a natural person in the course of a purely personal or household activity" (Article 2(2)(c) GDPR), meaning with no connection to a professional or commercial activity, but it does apply to those providing the means (recital 18 GDPR). In our reading both exemptions point the same way: once the use is work-related, they fall away.
Paragraph 12: free and open-source licences
The Regulation does not apply to AI systems released under free and open-source licences, "unless they are placed on the market or put into service as high-risk AI systems or as an AI system that falls under Article 5 or 50". For general-purpose AI models a narrower exception applies in Article 53(2), which does not cover the copyright policy and the training content summary (recital 104) and does not apply to a model with systemic risk; see our piece on Article 53.
According to recital 103, AI components provided against a price or otherwise monetised, including through services or the use of personal data, do not benefit from the open-source exceptions; transactions between microenterprises and making components available through open repositories do not count. The recital speaks of components; in our reading the same idea applies to paragraph 12: those who share a system freely can rely on it, those who make its use accessible only against payment cannot. For models the Commission works this out in its guidelines, points 82 to 86.
What remains unaffected: paragraphs 5, 7, 9 and 11
- Paragraph 5. Chapter II of the Digital Services Act (Regulation (EU) 2022/2065) on the liability of providers of intermediary services continues to apply.
- Paragraph 7. Union law on the protection of personal data, privacy and the confidentiality of communications continues to apply. The Regulation does not affect the GDPR, Regulation (EU) 2018/1725 or Directives 2002/58/EC and (EU) 2016/680, without prejudice to Article 4a and Article 59. Before the omnibus this referred to Article 10(5) of the original text; that paragraph has been deleted and the rule now sits in Article 4a (our piece).
- Paragraph 9. Rules on consumer protection and product safety continue to apply.
- Paragraph 11. The Union and Member States may maintain or introduce rules more favourable to workers on the use of AI systems by employers, and allow more favourable collective agreements.
Dates
Article 2 is in Chapter I and has applied since 2 February 2025 (Article 113, third paragraph, point (a)). The current wording of paragraphs 2 and 7 and the new paragraph 13 have applied since 27 July 2026, when the omnibus entered into force (Article 1, points (2) and (3), of Regulation (EU) 2026/1744). Which obligation applies is a separate question: Article 50 since 2 August 2026 (paragraph 2 for systems already placed on the market before then: by 2 December 2026, Article 111(4)), the requirements for high-risk AI systems under Annex III from 2 December 2027 and under Annex I from 2 August 2028 (Article 113, third paragraph, point (c)).
The municipality: three situations
Chatbot from an American supplier. The supplier is a provider (paragraph 1, point (a)) and, supplying directly, in our reading puts the system into service in the Union; a Dutch party selling it under the supplier's name is an importer (point (d)). The municipality is a deployer (point (b)) and takes measures on AI literacy (Article 4). A chatbot for general questions is as a rule not a high-risk AI system and so needs no authorised representative (the supplier's own underlying general-purpose AI model may, under Article 54); Article 50(1) does apply.
Camera system for public order. In our reading it falls under the AI Act: public order is not national security and the system is not used exclusively for it. If it identifies people remotely, it falls under point 1(a) of Annex III. The prohibition in Article 5(1)(h) covers "real-time" remote biometric identification in publicly accessible spaces for the purposes of law enforcement, that is by or on behalf of an authority competent for criminal offences and public security (Article 3, points (45) and (46)). Whether the municipality acts in that capacity depends, in our reading, on who operates the system and for what purpose. If not, Article 9(1) GDPR prohibits processing biometric data for unique identification, subject to limited exceptions (recital 39); Article 29 of the Dutch GDPR Implementation Act only allows it for authentication or security.
Employee using a chatbot at home. Looking up a recipe is purely personal; a letter to a resident is work, also with a private account. In our reading the municipality is then the deployer, since the employee acts on its behalf and Article 4 also covers staff using AI on its behalf. It is the controller for the personal data; with a private account there is no processing agreement with the chatbot provider (Article 28 GDPR), so residents' personal data do not belong there.
Bank or insurer
A bank has the creditworthiness of individuals assessed by a fintech in the United Kingdom, which runs the model on its own servers and sends back only scores. This is the case of recital 22: the system is not placed on the market, put into service or used in the Union, but the fintech falls under paragraph 1, point (c). It is in principle high-risk from 2 December 2027 (point 5(b) of Annex III; for an existing system Article 111(2)). Whether the bank is a deployer depends on who uses the system under its authority (Article 3, point (4)); in our reading it is if it enters the applications itself and decides on the use. An authorised representative is required before making the system available on the Union market (Article 22(1)); with output only, in our reading that does not arise, unlike when the bank deploys the system itself. Settle the roles by contract.
An insurer that adapts an open-source model for risk assessment and pricing in life and health insurance (point 5(c) of Annex III) and puts it into service under its own name is both provider and deployer. Paragraph 12 does not help: the system is put into service as a high-risk AI system.
For their workers, national law for which paragraph 11 in our reading leaves room applies, such as the consent right of the works council for facilities for monitoring attendance, conduct or performance (Article 27(1)(l) of the Dutch Works Councils Act). For a high-risk AI system at the workplace, the information duty of Article 26(7) is added.
Which AI Act and GDPR obligations apply to your case, you can see with the case check.
Frequently asked questions about Article 2
Sources
Newsletter
Every Tuesday, the AI Act week ahead in 5 minutes
A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.
Practical and short · No spam · One-click unsubscribe