Skip to main content
Praxikon

Fines on Union institutions: what Article 100 of the AI Act provides

··14 min read

Does your organisation supply an AI system to an EU agency, or share one with a European body? Then supervision lies partly outside the Netherlands. For Union institutions, bodies, offices and agencies the European Data Protection Supervisor (EDPS) is the supervisor, and Article 100 gives it a fining regime of its own. The Digital Omnibus (Regulation (EU) 2026/1744) did not amend Article 100.

In brief

The EDPS may impose an administrative fine on Union institutions, bodies, offices and agencies: up to EUR 1 500 000 for a prohibited AI practice under Article 5 (paragraph 2) and up to EUR 750 000 for non-compliance of the AI system with other requirements or obligations under the AI Act (paragraph 3). There is no turnover percentage; the annual budget of the institution is one of the criteria. The institution is heard and has access to the file. Article 100 fines the institution, not its supplier: the supplier falls under Article 99 and the national rules.

Who is fined, and who is not

Article 100(1) is addressed to "Union institutions, bodies, offices and agencies falling within the scope of this Regulation", for example the Commission or an EU agency, as provider or deployer. The definitions of those roles expressly include a public authority, agency or other body (Article 3, points (3) and (4)).

The supplier is not covered. A Dutch company that supplies an AI system to an agency under its own name or trademark is usually the provider (Article 3, point (3)). Its obligations are subject to the penalties the Member State lays down under Article 99. In the Netherlands those rules do not yet exist: the draft AI Act Implementation Act (Uitvoeringswet AI-verordening) is not yet law.

The EDPS itself states that it is not competent for complaints concerning private entities; those go to a national market surveillance authority (Complaints under the AI Act). In our reading, it may still request information as market surveillance authority about a system an institution uses (Article 3, point (48)). The same system can therefore face two supervisors and two fining regimes, depending on who breaches which obligation.

The EDPS as supervisor

For AI systems put into service or used by Union institutions, references to national competent authorities or market surveillance authorities are construed as references to the EDPS (Article 3, point (48)). It is their competent authority (Article 70(9)) and their market surveillance authority, except in relation to the Court of Justice of the European Union acting in its judicial capacity (Article 74(9)). Recital 168 states that it should have the power to impose fines on them.

The EDPS is preparing for this. In its Compass for its new role under the AI Act of 17 March 2026 it writes that it set up an AI unit in October 2024 and intends to lay down enforcement procedures in 2026 and 2027, including on handling complaints and applying administrative fines in accordance with Article 100.

Two maximums, no turnover

Article 99 (Member States)Article 100 (EDPS)
Prohibited practice (Article 5)Up to EUR 35 000 000 or 7 % of worldwide annual turnover (paragraph 3)Up to EUR 1 500 000 (paragraph 2)
Other requirements or obligationsUp to EUR 15 000 000 or 3 %, for the list in paragraph 4Up to EUR 750 000 (paragraph 3)
Incorrect information in reply to a request from an authority or notified bodyUp to EUR 7 500 000 or 1 % (paragraph 5)No tier of its own

Three differences stand out. The percentage in Article 99 applies only to an undertaking; Article 100 has fixed amounts only. Article 99(4) contains a closed list of obligations, whereas Article 100(3) refers to the "non-compliance of the AI system with any requirements or obligations under this Regulation". In our reading, that tier is not limited to a list. Whether it also covers obligations that do not attach to the system itself, such as AI literacy (Article 4) or the obligations of a deployer, is not yet settled. And Article 100 has no separate tier for incorrect information.

How the EDPS weighs and decides

When deciding whether to fine and on the amount, the EDPS takes all relevant circumstances into account. Paragraph 1 lists seven: the nature, gravity and duration of the infringement and of its consequences, the degree of responsibility of the institution, any action it took to mitigate damage, its cooperation with the EDPS (including compliance with measures previously ordered with regard to the same subject matter), any similar previous infringements, whether it notified the infringement itself, and its annual budget. Where Article 99(7) looks at turnover, market share and financial benefits, Article 100 looks at the budget. The fine must not affect the effective operation of the institution (paragraph 6).

Before the EDPS decides, the institution is given the opportunity of being heard. The EDPS bases its decision only on elements and circumstances on which the parties concerned have been able to comment, and complainants are associated closely with the proceedings (paragraph 4). The parties have access to the file, subject to the legitimate interest of individuals or undertakings in the protection of their personal data or business secrets (paragraph 5). In our reading, that reservation matters to a supplier: technical documentation in the file may contain business secrets.

Article 100 itself does not provide for an appeal. Unlike Article 101(5), for Commission fines on model providers, it does not mention unlimited jurisdiction of the Court of Justice. The money goes to the general budget of the Union (paragraph 6), and the EDPS notifies the Commission each year of the fines it has imposed and the proceedings it has initiated (paragraph 7).

Since when

Chapter XII, which contains Article 100, has applied since 2 August 2025; Article 101 only since 2 August 2026 (Article 113, second paragraph and third paragraph, point (b)). A fine is only possible for an obligation that itself already applies:

  • AI literacy and prohibited practices (Articles 4 and 5): since 2 February 2025; the new prohibitions in Article 5(1), points (ba) and (bb), and paragraphs 1a and 1b from 2 December 2026.
  • Transparency (Article 50): since 2 August 2026.
  • High-risk obligations: from 2 December 2027 for Annex III and from 2 August 2028 for Annex I. For a system placed on the market or put into service before then, they apply only after significant changes in its design; systems intended to be used by public authorities must comply by 2 August 2030 at the latest (Article 111(2)). Components of the large-scale IT systems in Annex X that were placed on the market or put into service before 2 August 2027 must comply by 31 December 2030 (Article 111(1)).

In our reading, the EDPS needs no national implementing act: Article 100 designates it and sets the amounts itself. The draft explanatory memorandum to the Dutch AI Act Implementation Act likewise lists Article 100 as needing no implementation by its nature.

Comparison with the privacy rules and with the Dutch public sector

Under Regulation (EU) 2018/1725, the data protection law for the Union institutions, the EDPS may also impose fines. It may do so where an institution fails to comply with an order under Article 58(2) of that Regulation, points (d) to (h) and (j) (Article 66 of Regulation (EU) 2018/1725). The maximums are EUR 25 000 per infringement and EUR 250 000 per year, or EUR 50 000 per infringement and EUR 500 000 per year for, among other things, the principles and the rights of data subjects. Article 100 of the AI Act mentions no prior order and has higher maximums per infringement; compliance with measures previously ordered is, however, one of the criteria (paragraph 1, point (d)).

For national public bodies, each Member State lays down rules on to what extent administrative fines may be imposed on public authorities and bodies established in that Member State (Article 99(8)). The Netherlands has not yet done so; according to its explanatory memorandum, the draft Implementation Act does not rule out fines on public bodies. More on this in what are the fines under the AI Act, and can a municipality get one?. For the Union institutions the choice has already been made: they can be fined, up to lower fixed amounts.

Two examples

An executive agency shares a system. Suppose a Dutch executive agency has built an AI system that sorts incoming documents and shares it with an EU agency, which uses it under its own authority. The example is fictional.

  1. The EU agency is the deployer and falls under the EDPS. If it breaches an obligation that already applies, the EDPS can fine it up to EUR 1 500 000 for a prohibited practice and, in our reading, up to EUR 750 000 where the breach falls under paragraph 3.
  2. The Dutch organisation is, in our reading, the provider if it places the system on the market or puts it into service under its own name, whether for payment or free of charge (Article 3, point (3)). It is subject to Article 99 and the Dutch choice under paragraph 8, which has not yet been made.
  3. High-risk? If the system falls under Annex III, it is in principle high-risk, subject to Article 6(3). From 2 December 2027 the EU agency must then register its use (Article 26(8), with Article 49(3)) and, in our reading, perform a fundamental rights impact assessment before use (Article 27). If the system was put into service before then, this applies only after significant changes in its design and, in our reading, by 2 August 2030 at the latest (Article 111(2)). The same applies to the Dutch organisation if it uses the system itself.

A bank as supplier. Suppose a Dutch bank sells a customer chatbot to an EU agency through a subsidiary. If the subsidiary supplies the chatbot under its own name or trademark, it is, in our reading, the provider and must ensure that people are informed that they are interacting with an AI system (Article 50(1)). If it falls short, the tier of Article 99(4), point (g), applies, with the national penalty rules the Netherlands has yet to lay down; the EDPS does not fine it. This example is fictional too.

Has a fine been imposed under Article 100 yet?

On 19 November 2026 we found no published fining decision under Article 100. We searched the EDPS page on its work under the AI Act and the overview of EDPS press releases. That is no proof that no decision exists. We wrote about the EDPS's earlier work on AI in the revision of the GenAI guidance.

What to do now

  1. If you work with or for a Union institution, record for each AI system who is the provider and who is the deployer. That determines which supervisor and which fining regime applies.
  2. Agree in the contract how you cooperate in an EDPS investigation, and which documentation you designate as a business secret.
  3. Check which obligations already apply: Article 4 (AI literacy), Article 5 and Article 50. The high-risk obligations follow later.

For more on the powers of market surveillance authorities, see market surveillance powers.

Frequently asked questions about Article 100 of the AI Act

Sources

EUR-Lex: Regulation (EU) 2024/1689 (AI Act), original text, recital 168 (13 June 2024, accessed 19 November 2026)
EUR-Lex: Regulation (EU) 2026/1744 (Digital Omnibus on AI) (8 July 2026, published 24 July 2026, accessed 19 November 2026)
European Data Protection Supervisor: Towards trustworthy AI in the EU public administration: The EDPS Compass for its new role under the AI Act 2026-2027 (17 March 2026, accessed 19 November 2026)
European Data Protection Supervisor: Complaints under the AI Act (accessed 19 November 2026)
European Data Protection Supervisor: Our work under the AI Act (accessed 19 November 2026)
European Data Protection Supervisor: EDPS Press Releases (accessed 19 November 2026)

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

Written by

Zahed Ashkara, freelance AI and privacy consultant and legal adviser

Zahed Ashkara

Freelance AI and privacy consultant and legal adviser · AI governance & safety

Zahed Ashkara is a freelance AI and privacy consultant and legal adviser for the GDPR and the EU AI Act. Organisations hire him as AI governance consultant, privacy counsel, external DPO or AI governance lead, have him carry out a DPIA, FRIA or AI Act classification, or book an in-company training. He focuses on AI governance and safety, including AI agents. He works exclusively as a freelancer, through Embed AI (trade name Praxikon). He is a Certified AI Compliance Officer (CAICO) and has been a member of the NEN standards committee Artificial Intelligence & Big Data since July 2026.

Follow on LinkedInView expert profile

Use these sources in your AI assistant too: Praxikon in Claude and other assistants

Real-World Stories

Discover how organizations apply this in practice

Related articles

What if you do not comply with Article 50: enforcement and fines since 2 August 2026

The transparency obligations under Article 50 are enforceable since 2 August 2026 and were not postponed by the Digital Omnibus. A breach can lead to a fine of up to 15 million euro or 3 percent of total worldwide annual turnover, whichever is higher (lower for SMEs, start-ups and small mid-caps). Enforcement generally sits with the national market surveillance authorities. The duty depends on your role: providers disclose AI interaction and mark synthetic output, deployers inform about emotion recognition and disclose deepfakes and certain AI text; build an evidence layer for it.

EU AI ActArticle 50Enforcement

When the regulator knocks: Articles 79 to 82 of the AI Act on AI systems presenting a risk

What happens when a market surveillance authority considers that an AI system presents a risk? Article 79 sets out the national procedure: evaluation, corrective action within a period of at most 15 working days, and where needed a prohibition or recall. Article 80 deals with a system the provider wrongly classified as non-high-risk, Article 81 with the Union safeguard procedure and Article 82 with a system that complies with the Regulation and still presents a risk. With who will supervise in the Netherlands, from when it applies, and a municipality using a system for social assistance applications and a bank as examples.

EU AI ActArticle 79Article 80