Article 99 penalties: the three tiers, and which one your hiring tool sits in

Edition 13. A reading time of about 5 minutes.

The hook

You have probably seen the headline number: fines of up to 35 million euro, or 7 percent of global turnover, for breaking the EU AI Act. It is real, and it is meant to be felt. What gets lost is that this is the top of three separate tiers, and it does not apply to most AI hiring tools. The 35 million euro ceiling is reserved for the AI practices the Act bans outright. The obligations that govern a high-risk hiring system sit one tier down. Knowing which tier you are in changes what the risk actually is, and that is worth getting right before you let a headline set your budget.

What this means

Article 99 sets administrative fines in three bands, each a ceiling rather than an automatic charge, and each applied by national authorities, not by Brussels (AI Act, Article 99).

The top band is for the prohibited practices. Breaking the ban on the AI practices listed in Article 5, the uses the Act forbids entirely, carries fines "of up to 35 000 000 EUR or, if the offender is an undertaking, up to 7 % of its total worldwide annual turnover for the preceding financial year, whichever is higher" (Article 99(3)). This is the number in the headlines. For a standard recruiting or screening tool, it is not the number that applies.

The middle band is the one that governs high-risk hiring. Non-compliance with the operator obligations, the duties on providers, deployers, importers, distributors and notified bodies, plus the Article 50 transparency duties that bind providers and deployers, draws fines "of up to 15 000 000 EUR or, if the offender is an undertaking, up to 3 % of its total worldwide annual turnover for the preceding financial year, whichever is higher" (Article 99(4)). If your AI hiring system is classed as high-risk and you fall short on, say, human oversight or record-keeping, this is the tier you are in.

The lowest band targets a specific behaviour: misleading the regulator. Supplying "incorrect, incomplete or misleading information to notified bodies or national competent authorities in reply to a request" carries fines of up to 7 500 000 EUR or 1 percent of worldwide annual turnover, whichever is higher (Article 99(5)). The lesson there is plain: when an authority asks, the honest, complete answer is also the cheap one.

Two qualifiers matter for mid-market teams. First, smaller companies are treated proportionately: for SMEs, including start-ups, each fine is capped at "the percentages or amount referred to in paragraphs 3, 4 and 5, whichever thereof is lower" (Article 99(6)). For a smaller business, the percentage of turnover is usually the lower figure, so that, not the multimillion ceiling, is the cap. Second, every figure is an upper limit. The text says "up to" in each tier, and national authorities decide the actual amount.

Timing is the part that needs care, because the Act's calendar has moving pieces. The penalties chapter itself, Chapter XII, has applied since 2 August 2025 (AI Act, Article 113). But a fine can only attach to an obligation that is itself in force. The prohibitions in Article 5 have applied since 2 February 2025, so the top tier already has teeth. The high-risk hiring obligations that the middle tier enforces flow from Annex III, the classification route in Article 6(2), and under the Act as published they apply with the general application date of 2 August 2026. Honest caveat: under the Commission's provisional Digital Omnibus agreement, Annex III employment systems would move to 2 December 2027, but that agreement is not yet in the Official Journal, so treat the new date as provisional. We will flag it here the moment it is settled. One trap worth naming: the 2 August 2027 date you may see quoted elsewhere belongs to Article 6(1), the separate high-risk route for AI embedded in regulated products under Annex I. It is not the hiring date, and planning around it would leave you a year late under the law as it stands.

What to do with this

You do not need a lawyer to take the first steps, and you should not let the largest number drive decisions it does not govern. Three moves.

1. Place yourself in the right tier before you size the risk. Confirm whether anything you run touches a prohibited practice under Article 5, that is the only path to the 35 million euro band. If not, and your tool is high-risk, your exposure runs through the 15 million euro or 3 percent tier, and if you are an SME the lower-of cap applies. Sizing risk against the wrong tier wastes budget and attention.
2. Treat the "misleading information" tier as a process rule, not a legal one. Decide now who answers a regulator's request, and make completeness and accuracy the standard. A rushed or partial reply can create a separate, avoidable fine on top of whatever prompted the question. Keep your evidence trail current so the accurate answer is easy to give.
3. Map your obligations to their live dates, not to the penalty date. The penalty framework is already on the books, but it bites only as each underlying obligation comes into force. Build your timeline around when your high-risk duties actually apply, and keep one eye on the Digital Omnibus, which could shift them.

Read this way, Article 99 is less a threat than a map. The ceilings are deliberately high to make the rules matter, but the band that applies to you, the proportionality for smaller firms, and the "up to" in every tier mean the real exposure is more measured than the headline suggests. The teams that know exactly which tier they sit in are the ones who spend their compliance effort where it counts.

This is educational material and a starting point, not legal advice. The EU AI Act is still being amended, and parts of the high-risk timetable are among the provisions proposed for change under the Commission's Digital Omnibus package, which is not yet final law. How a fine would be assessed depends on national implementing rules and the specific facts, and Member States set their own penalty regimes within these ceilings. For your situation, consult a qualified lawyer. Signato is not a law firm and does not certify compliance.

Not sure which tier your hiring stack falls in? Take the free AI Hiring Risk Self-Assessment to see where you are exposed, then see the same evidence-first discipline applied to what your AI sends out: the free local trial of the Signato chamber blocks a draft that pairs the wrong person with the wrong subject before it leaves, no card, nothing leaves your machine.

Every claim here is traced to the primary regulation and checked by a person before it goes out. We tell you what is settled, what is still open, and what to do next. Know someone who uses AI in hiring in the EU? Forward this to them.

Sources (primary, read 2026-06-08):

- AI Act, Article 99 (Penalties), the three tiers: Article 99(3) up to 35 000 000 EUR or 7 percent of worldwide annual turnover for Article 5 prohibited practices; Article 99(4) up to 15 000 000 EUR or 3 percent for other operator obligations including providers, deployers and Article 50 transparency; Article 99(5) up to 7 500 000 EUR or 1 percent for incorrect, incomplete or misleading information to authorities; Article 99(6) SMEs and start-ups capped at whichever is lower. - AI Act, Article 113 (Entry into force and application), Chapter XII penalties apply from 2 August 2025 (except Article 101); Article 5 prohibitions from 2 February 2025; general application 2 August 2026, which carries the Annex III / Article 6(2) high-risk obligations, including hiring; Article 6(1), the high-risk route for AI embedded in Annex I regulated products, and its corresponding obligations from 2 August 2027. - AI Act, Article 6 (Classification rules for high-risk AI systems), Article 6(2) routes Annex III systems, including employment and worker management, into the high-risk class; Article 6(1) covers AI embedded in products under Annex I. - Canonical text: Regulation (EU) 2024/1689, EUR-Lex CELEX:32024R1689.