HomeAI RegulationEU AI Act Fines in 2026: Tiers, Examples and Who Pays

EU AI Act Fines in 2026: Tiers, Examples and Who Pays

Last updated: September 29, 2026 · Figures checked against the consolidated AI Act text (as amended by Regulation (EU) 2026/1744) and the Commission’s enforcement page on this date

The EU AI Act has three fine tiers: up to €35 million or 7% of worldwide annual turnover for prohibited AI practices, €15 million or 3% for most other breaches, and €7.5 million or 1% for misleading authorities. Large companies pay whichever figure is higher. SMEs and start-ups pay whichever is lower, and since July 2026 small mid-caps get the same relief for the two lower tiers. The Commission can fine general-purpose AI model providers up to €15 million or 3% from August 2, 2026. As of September 29, 2026, no AI Act fine has been published.

Article 99–101Omnibus 2026/1744No fines issued yetChecked Sep 29, 2026

Most pages about AI Act penalties stop at the three numbers. We went further. Before writing, we compared eight of the top-ranking articles on this topic question by question. None worked through a real calculation, none said whether a company can be fined under both GDPR and the AI Act, none covered appeals, and only one said plainly that nobody has been fined yet. This guide covers those gaps and the July 2026 changes from the Digital Omnibus. For the wider regulation, start with our EU AI Act explainer.

What are the EU AI Act fine tiers?

Article 99 sets three ceilings. Each has a fixed euro amount and a percentage of the company’s total worldwide annual turnover for the preceding financial year.

TierWhat it coversMaximum fineLegal basis
1 — Prohibited practicesBanned AI uses listed in Article 5: manipulative or exploitative systems, social scoring, untargeted scraping of facial images, emotion recognition at work or school, certain biometric categorisation and real-time remote biometric identification. From December 2, 2026 the list also covers AI that generates non-consensual intimate content or child sexual abuse material.€35M or 7%Art. 99(3)
2 — Operator obligationsDuties of providers (Art. 16), authorised representatives (22), importers (23), distributors (24), deployers (26), notified bodies (31, 33, 34), Art. 25 value-chain duties, and the transparency rules in Article 50€15M or 3%Art. 99(4)
3 — Misleading authoritiesIncorrect, incomplete or misleading information given to notified bodies or national authorities in reply to a request€7.5M or 1%Art. 99(5)
GPAI model providersBreaches by providers of general-purpose AI models, fined by the Commission, not national authorities€15M or 3%Art. 101
EU institutionsUnion bodies, fined by the European Data Protection Supervisor€1.5M (prohibited) / €750,000 (other)Art. 100

The percentage is taken from worldwide turnover, not EU revenue and not the revenue of the product at fault. The Act borrows the competition-law idea of an “undertaking”, which usually means the whole economic group rather than the one subsidiary that broke the rules.

Is the fine the higher or the lower of the two numbers?

It depends on the size of the company:

  • Large companies: whichever is higher. For a group with billions in turnover, the percentage decides.
  • SMEs and start-ups: whichever is lower, in all three tiers (Art. 99(6)).
  • Small mid-caps (new since July 27, 2026): whichever is lower, but only in tiers 2 and 3 (Art. 99(6a)). For prohibited practices they are treated like large companies. A small mid-cap under Commission Recommendation (EU) 2025/1099 is a company that isn’t an SME, has fewer than 750 employees, and has turnover of up to €150 million or a balance sheet of up to €129 million.
Which EU AI Act fine cap applies to your company Decision flow: SMEs and start-ups pay the lower of the euro amount and the turnover percentage in all three tiers; small mid-caps pay the lower figure only for the €15M/3% and €7.5M/1% tiers; all other companies pay the higher figure, up to €35M or 7% for prohibited practices. EU AI Act fine cap decision flowDecodeTheFuture.orgeu ai act fines, ai act penalties, article 99, sme, small mid-capHow company size decides whether the higher or lower AI Act fine cap applies after the Digital Omnibus 2026/1744.Diagramimage/svg+xmlen© DecodeTheFuture.org Who is being fined? SME or start-up? Yes → LOWER figure in all 3 tiers (Art. 99(6)) no Small mid-cap? (<750 staff, ≤€150M) Yes → LOWER figure for €15M/3% and €7.5M/1% but HIGHER for prohibited practices (Art. 99(6a)) no Large company HIGHER figure in every tier Caps (worldwide turnover) Prohibited: €35M or 7% Most obligations: €15M or 3% Misleading info: €7.5M or 1%

Worked examples: what is the maximum fine for your company?

These are the ceilings from Article 99, calculated for four made-up companies. Real fines are set case by case and will usually be lower (see the factors below).

Company (worldwide turnover)Prohibited practiceMost obligationsMisleading information
Large group, €2 billion — higher figure applies7% = €140M3% = €60M1% = €20M
Large company, €300 million (above the small mid-cap threshold) — higher figure€35M (7% would be €21M)€15M (3% = €9M)€7.5M (1% = €3M)
Small mid-cap, €100 million, 500 staff€35M (no relief for this tier)lower: 3% = €3Mlower: 1% = €1M
SME, €20 millionlower: 7% = €1.4Mlower: 3% = €600,000lower: 1% = €200,000

Two things stand out. For a mid-sized company that isn’t an SME, the fixed amount is the real cap: a €300 million business faces the full €35 million for a prohibited practice even though 7% of its turnover is less. And the Omnibus change matters a lot in practice. A €100 million small mid-cap’s worst case for an ordinary obligation fell from €15 million to €3 million.

Has anyone been fined under the AI Act yet?

No. As of September 29, 2026, neither the European Commission nor any national authority has published an AI Act fine. The Commission’s enforcement page (updated August 24, 2026) lists none. Its enforcement powers over general-purpose AI models only started on August 2, 2026, and investigations usually take months.

Beware of invented fines. Several websites describe AI Act fines of “€47 million” or “€85 million in March 2026”. They name no company and link no decision. We found no official record of any of them. If a page cites an AI Act fine, check for the authority’s decision before you repeat it.

There’s also a practical reason for the silence. By the August 2025 deadline, many Member States had not fully designated their authorities. A June 2026 tracker counted 9 countries with a clear set-up, 12 with a partial one and 6 still unclear. Poland passed its implementing act in July 2026, but its fine procedures only apply from October 28, 2026.

When do AI Act penalties apply? Key dates

DateWhat changes for enforcement
February 2, 2025Prohibited practices (Art. 5) and AI literacy (Art. 4) apply
August 2, 2025Penalty chapter, GPAI obligations and governance apply; Member States had to notify penalty rules
July 27, 2026Digital Omnibus (Regulation 2026/1744) in force: small mid-cap relief, new Art. 99(4)(da)
August 2, 2026Commission can fine GPAI providers (Art. 101); Article 50 transparency applies; national enforcement powers apply
October 28, 2026Poland: fine and inspection procedures of the national AI act apply
December 2, 2026New bans on AI-generated non-consensual intimate content and CSAM; end of the grace period for marking AI-generated content (Art. 50(2)) for systems placed on the market before August 2, 2026
August 2, 2027GPAI models placed on the market before August 2, 2025 must comply
December 2, 2027High-risk obligations for Annex III systems (moved from August 2026)
August 2, 2028High-risk obligations for AI in regulated products (Annex I)

Who imposes AI Act fines?

  • National market surveillance authorities fine providers, deployers, importers, distributors and others under Article 99. In Poland that is the new Komisja Rozwoju i Bezpieczeństwa Sztucznej Inteligencji (KRiBSI).
  • The European Commission (through the AI Office) fines providers of general-purpose AI models under Article 101.
  • The European Data Protection Supervisor fines EU institutions, bodies and agencies under Article 100.

Each Member State writes its own procedure. Article 99(8) also lets each country decide whether, and how much, its own public bodies can be fined.

Can deployers be fined, or only AI developers?

Deployers can be fined too. A deployer is any organisation that uses an AI system in a professional capacity, even if someone else built it. Breaching the deployer duties in Article 26 falls in the €15 million / 3% tier. For a high-risk system those duties include using it according to the instructions, assigning human oversight, monitoring it and keeping logs. The same tier covers the transparency duties that fall on deployers under Article 50, such as disclosing deepfakes and AI-generated text published to inform the public. Purely personal, non-professional use is outside the Act.

How are fines set? The factors that raise or lower a fine

Article 99(7) tells authorities to weigh, case by case:

  1. the nature, gravity and duration of the breach, the system’s purpose, and how many people were affected and harmed;
  2. whether other market surveillance authorities already fined the same operator for the same breach;
  3. whether fines were already imposed under other EU or national law for the same act;
  4. the operator’s size, annual turnover and market share;
  5. other aggravating or mitigating factors, such as money gained or losses avoided;
  6. how well the operator cooperated with the authorities;
  7. the operator’s degree of responsibility, given its technical and organisational measures;
  8. how the authority learned of the breach, including whether the operator reported it;
  9. whether it was intentional or negligent;
  10. what the operator did to reduce the harm.

What this means in practice: documentation is your best mitigation. Factors 6, 7, 8 and 10 all reward a company that can show a working risk process, prompt self-reporting and a fix. A company that finds a problem and has nothing on paper loses most of those arguments. Our comparison of AI risk management frameworks shows what “on paper” looks like.

Can you be fined under both GDPR and the AI Act?

Yes, and nothing in the Act prevents it. Many AI systems process personal data, so the same incident can breach both laws. For example, a recruitment tool might profile candidates without a lawful basis and also skip the deployer’s human-oversight duty. The two regimes are enforced by different authorities, and the AI Act’s own ceilings are higher: €35 million / 7% against GDPR’s €20 million / 4%.

The Act softens the double hit instead of banning it. Factor (c) in Article 99(7) requires the authority to take into account fines already imposed under other law for the same act. Expect the second fine to be lower, not cancelled. The Act doesn’t settle the details of how the two will interact, so watch the first cases.

Are managers personally liable under the AI Act?

Not under the Regulation itself. Article 99 fines “operators”, meaning the company acting as provider, deployer, importer or distributor. It creates no personal fine for directors or employees. Two caveats:

  • A self-employed person using AI professionally is an operator and can be fined directly.
  • Member States may add their own penalties and “other enforcement measures” (Art. 99(1)), and national company and criminal law still apply. Poland’s act, for example, has a separate chapter of criminal provisions. Check your national law. The AI Act alone doesn’t answer this question.

What sanctions exist besides fines?

For many companies the bigger risk isn’t the fine but losing the product. Under the market surveillance procedure (Art. 79), an authority that finds a non-compliant or risky AI system can order the operator to fix it, withdraw it from the market or recall it within a set period. If the operator doesn’t comply, the authority can restrict or prohibit the system itself. For general-purpose AI models, the Commission can require mitigation measures, restrict availability, or order a withdrawal or recall (Art. 93). Article 99(1) also allows warnings and non-monetary measures.

Can AI Act fines be appealed?

Yes. Article 99 requires the authorities’ powers to be subject to procedural safeguards, “including effective judicial remedies and due process”. The route depends on who fined you:

  • National fines: through national courts. In Poland, KRiBSI decisions are appealed to the Regional Court in Warsaw’s competition and consumer protection court (SOKiK), filed through KRiBSI.
  • Commission fines on GPAI providers: before the EU courts, like other Commission decisions.

Poland’s act adds two incentives that none of the eight articles we compared mention. A fine can be cut by 10–50% if the operator carries out a KRiBSI warning within three months, and by 20–70% through a settlement with the authority. Euro caps are converted to złoty at the NBP average rate of January 28 each year.

Is there a fine for ignoring the AI literacy obligation?

Not a dedicated one. Article 4 requires providers and deployers to ensure a sufficient level of AI literacy among their staff, and it has applied since February 2, 2025. But Article 4 isn’t among the provisions listed in the Article 99 fine tiers. Poor staff training is more likely to show up in a fine for something else: for example, as a weak point in factor (g), “degree of responsibility”, when a deployer’s untrained staff misuse a high-risk system.

Does the AI Act apply to non-EU companies?

Yes. The Act follows the market, not the head office. A US or UK provider that places an AI system on the EU market, or whose system’s output is used in the EU, falls under it and can be fined. The percentage is still calculated on worldwide turnover. Non-EU providers of high-risk systems and GPAI models must appoint an authorised representative in the EU, and the authorised representative’s own duties are in the €15 million / 3% tier.

What should a company do now?

  1. Screen for prohibited practices first. They have applied since February 2025, carry the highest fines, and give small mid-caps no relief. Add the new December 2026 bans to the screen.
  2. Classify your size correctly. SME, small mid-cap or large company changes your maximum exposure by an order of magnitude (see the table above).
  3. Fix Article 50 transparency now. Chatbot disclosure and deepfake labelling have applied since August 2, 2026, while most high-risk duties don’t start until December 2027.
  4. Build the paper trail. The Article 99(7) factors reward documented risk management, cooperation and self-reporting. See our AI compliance workflows guide.
  5. Map your GDPR exposure alongside. The same incident can trigger both regimes.

FAQ

What is the maximum fine under the EU AI Act?

€35 million or 7% of total worldwide annual turnover, whichever is higher, for prohibited AI practices under Article 5. SMEs and start-ups pay whichever is lower.

Has any company been fined under the AI Act?

No. As of September 29, 2026, no AI Act fine has been published by the European Commission or any national authority. Reports of specific multimillion fines circulating online are not backed by any official decision.

How much can a small company be fined under the AI Act?

SMEs and start-ups pay the lower of the two figures in every tier. An SME with €20 million turnover faces at most €1.4 million for a prohibited practice, €600,000 for most other breaches and €200,000 for misleading information.

What changed in AI Act fines with the Digital Omnibus?

Regulation (EU) 2026/1744, in force since July 27, 2026, extended the lower-of-two rule to small mid-caps for the €15 million and €7.5 million tiers, added Article 25 value-chain duties to the middle tier, created new prohibited practices from December 2, 2026, and moved high-risk obligations to December 2027 and August 2028.

Who fines providers of general-purpose AI models like GPT or Claude?

The European Commission, through the AI Office, under Article 101: up to €15 million or 3% of worldwide turnover, whichever is higher. Those powers apply from August 2, 2026.

Can a company be fined under GDPR and the AI Act for the same thing?

Yes. The regimes are separate, but Article 99(7)(c) requires authorities to take into account fines already imposed under other law for the same act when setting the AI Act fine.

Who imposes AI Act fines in Poland?

The Komisja Rozwoju i Bezpieczeństwa Sztucznej Inteligencji (KRiBSI), under the Polish AI act (Dz.U. 2026 poz. 1003). Its fine procedures apply from October 28, 2026, and appeals go to the Regional Court in Warsaw (SOKiK).

Related: EU AI Act explained · NIST AI RMF vs ISO 42001 vs EU AI Act · AI agent security. Polish readers: kary za naruszenie AI Act — przewodnik po polsku.

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

- Advertisment -

Most Popular

Recent Comments