/

Updated 18 November 2026

EU AI Act Fines and Penalties Explained (2026)

AI Disclosure: This content includes AI-generated text.

Short answer: EU AI Act fines come in three tiers. Using a banned AI practice can cost up to 35 million euros or 7% of worldwide annual turnover. Breaching most other obligations, including the Article 50 transparency rules that matter for websites, can cost up to 15 million euros or 3%. Giving authorities false or misleading information can cost up to 7.5 million euros or 1%. For SMEs and start-ups, the lower of the two amounts applies, and every fine must be proportionate.

This guide explains the tiers, how authorities set the actual amount, who enforces the rules, and what the realistic risk looks like for a small website.

The three fine tiers (Article 99)

Breach Maximum fine Website examples
Prohibited AI practices (Article 5) 35 million euros or 7% of worldwide annual turnover, whichever is higher Manipulative AI that exploits vulnerable users and causes significant harm; untargeted scraping of facial images
Most other obligations, including Article 50 transparency 15 million euros or 3%, whichever is higher Chatbot that hides it’s an AI; unlabelled deepfake; unreviewed public-interest AI text without disclosure
Incorrect, incomplete or misleading information to authorities or notified bodies 7.5 million euros or 1%, whichever is higher False answers to a market surveillance authority’s request

SMEs and start-ups: for these, each cap is the lower of the fixed amount and the percentage, not the higher. A small business with 2 million euros turnover therefore faces a theoretical Article 50 maximum of 60,000 euros (3%), not 15 million.

Providers of general-purpose AI models, such as the large model makers, have a separate regime under Article 101 enforced by the Commission, with fines up to 15 million euros or 3%.

How the actual amount is decided

The figures above are ceilings. Article 99 requires authorities to consider all relevant circumstances, including:

  • the nature, gravity and duration of the breach, and its consequences;
  • how many people were affected and how badly;
  • whether other authorities already fined the same conduct;
  • the size, turnover and market share of the business;
  • any financial benefit gained or loss avoided;
  • whether the breach was intentional or negligent;
  • cooperation with the authority and steps taken to fix the problem;
  • how the authority found out, including whether you reported it yourself.

Member States can also provide for warnings and non-monetary measures. For a first, minor transparency gap that’s fixed quickly, an order to comply is a far more likely outcome than a large fine.

Who enforces it

Each EU country designates national market surveillance authorities to enforce the AI Act, and sets its own procedural rules for penalties. In Germany, the Bundesnetzagentur is the central authority for most of the Act. Anyone can lodge a complaint with a market surveillance authority if they believe the AI Act has been breached (Article 85). The European AI Office at the Commission supervises general-purpose AI models.

When do the penalties apply?

  • 2 February 2025: the bans on prohibited practices and the AI literacy duty started to apply.
  • 2 August 2025: the penalty provisions started to apply, and Member States had to have their rules in place.
  • 2 August 2026: Article 50 transparency obligations started to apply, so breaches since then can be fined. Commission fines for general-purpose AI model providers also became possible.
  • 2 December 2026: end of the grace period for machine-readable marking by providers of generative AI systems already on the market.
  • 2 December 2027 and 2 August 2028: high-risk obligations, delayed by the Digital Omnibus.

The realistic risk for a website

For a small business site, the most likely trouble isn’t a regulator opening an investigation on its own. It’s one of these:

  • A complaint from a visitor, journalist or competitor about an unlabelled deepfake or a chatbot that pretended to be human.
  • A competitor’s legal action. In some countries, including Germany, competitors can challenge breaches of market conduct rules through unfair competition law. Lawyers expect transparency duties like Article 50 to be tested this way.
  • Consumer law claims when AI content misleads customers, regardless of the AI Act.
  • Reputation: a screenshot of a “human” support agent that turns out to be a bot travels fast.

The cheapest protection is to fix the visible issues and keep a short record of your decisions. If an authority ever asks, a dated list showing you checked your chatbot, labelled the right images and reviewed or labelled your articles counts in your favour under the factors above.

Five steps to stay well clear of fines

  1. Run a quick AI audit of your website.
  2. Make sure every chatbot says it’s an AI at the first message. See chatbot compliance.
  3. Label realistic AI images that could pass as real.
  4. Review or label AI-written text on public-interest topics. See the Article 50 content test.
  5. Write down what you decided and when.

The free EU AI Act Ready plugin for WordPress covers steps 2 to 5 for most sites and exports a compliance report as your record. It runs on your own server and needs no account.

Frequently asked questions

Can a small website really be fined 15 million euros?

No. For SMEs the cap is the lower of 15 million euros and 3% of turnover, and the amount must be proportionate to the breach and the business.

Is there a fine for not having an AI literacy programme?

The AI Act doesn’t list Article 4 in its fine tiers directly. Member States set their own enforcement rules, and authorities can consider AI literacy when assessing other breaches. Treat it as a real duty.

Can I be fined for a third-party tool’s failure?

Only for your own obligations as a deployer, or as provider if you built the system. See responsibility for third-party AI tools.

Does the AI Act give people a right to compensation?

The AI Act itself doesn’t create a compensation claim. Other laws, such as product liability, GDPR and national civil law, may.