The Cost of Ignorance: A Breakdown of EU AI Act Penalties and Fines

Penalties & Enforcement · 10 min read · Updated March 2026

The EU AI Act’s penalty regime is the most severe in European digital regulation history — its top tier exceeds even GDPR’s headline figures. Yet many organisations still treat compliance as a future concern. This guide breaks down every penalty tier, maps violations to real-world scenarios, compares the structure to GDPR, and explains how enforcement will work in practice — so your organisation can make informed risk decisions.

The Numbers

€35M

or 7% of global turnover — Prohibited AI (Art. 5)

€15M

or 3% of global turnover — Most high-risk violations (Art. 6–49)

€7.5M

or 1% of global turnover — Incorrect or misleading information (Art. 99(4))

Table of Contents

The three-tier penalty structure
Tier 1 (€35M / 7%): Prohibited AI violations
Tier 2 (€15M / 3%): High-risk AI and GPAI violations
Tier 3 (€7.5M / 1%): Misleading information
How EU AI Act fines compare to GDPR
How penalties are calculated: the mitigating and aggravating factors
SME and startup rules: reduced penalties and proportionality
Beyond fines: other enforcement powers
Real-world penalty scenarios
FAQ

1. The Three-Tier Penalty Structure

Article 99 of the EU AI Act establishes three administrative fine tiers, each calibrated to the severity of the underlying violation. The Act follows EU regulatory convention: fines are expressed as the higher of a fixed cap or a percentage of global annual turnover — ensuring that penalties are proportionate to large, globally operating companies while remaining meaningful for smaller entities.

Tier 1 — Most Severe

Up to €35,000,000 or 7% of global annual turnover

Whichever is higher

Applies to: violations of Article 5 (prohibited AI practices). In force since 2 February 2025. Exceeds GDPR’s maximum tier.

Tier 2 — Standard

Up to €15,000,000 or 3% of global annual turnover

Whichever is higher

Applies to: violations of most high-risk AI obligations (Articles 6–49), GPAI obligations (Articles 51–56), national competent authority obstruction. In force from 2 August 2025 (GPAI) and 2 August 2026 (high-risk AI).

Tier 3 — Information Violations

Up to €7,500,000 or 1% of global annual turnover

Whichever is higher

Applies to: supplying incorrect, incomplete, or misleading information to national competent authorities or the EU AI Office. This tier exists because accurate regulatory reporting is a foundational compliance obligation.

The “higher of” rule in practice: For a company with €10 billion global annual turnover, the 7% rule yields a maximum Tier 1 fine of €700 million — far above the €35M fixed cap. The percentage clause is the operative constraint for large multinationals. For a startup with €5M annual revenue, the fixed cap of €35M far exceeds 7% of €5M (= €350K) — the percentage clause is more lenient, but the fixed cap does not apply as a floor: the fine is capped at the lower of the two.

2. Tier 1 (€35M / 7%): Prohibited AI Violations

The highest penalty tier applies exclusively to violations of Article 5 — the prohibited AI practices. These are absolute bans with no conformity pathway. The nine prohibited practices (including the newly added non-consensual explicit imagery prohibition from March 2026) all attract Tier 1 fines.

Article 5 Violation	In Force Since	Maximum Fine
Subliminal or manipulative AI (Art. 5(1)(a)&(b))	Feb 2025	€35M or 7%
Social scoring by public authorities (Art. 5(1)(c))	Feb 2025	€35M or 7%
Predictive policing based solely on profiling (Art. 5(1)(d))	Feb 2025	€35M or 7%
Facial image scraping from internet / CCTV (Art. 5(1)(e))	Feb 2025	€35M or 7%
Emotion inference in workplaces/education (Art. 5(1)(f))	Feb 2025	€35M or 7%
Biometric categorisation by sensitive characteristics (Art. 5(1)(g))	Feb 2025	€35M or 7%
Real-time biometric surveillance in public (Art. 5(1)(h))	Feb 2025	€35M or 7%
NEW: Non-consensual explicit imagery generation (Art. 5(1)(i))	Mar 2026	€35M or 7%

3. Tier 2 (€15M / 3%): High-Risk AI and GPAI Violations

The standard enforcement tier covers the broad range of obligations for high-risk AI systems and GPAI models. This is where the majority of enforcement actions will take place after August 2026.

High-Risk AI Violations (Art. 6–49)

Failing to complete conformity assessment (Art. 43)
Missing CE marking or EU database registration (Art. 48–49)
Inadequate or missing technical documentation (Art. 11)
Failing to implement human oversight mechanisms (Art. 14)
Failing to implement adequate risk management system (Art. 9)
Placing a non-conformant high-risk AI on the EU market
Failing to appoint an Authorised Representative (Art. 22)
Deployer failing to conduct FRIA where required (Art. 27)
Failing to notify serious incidents within 15 days (Art. 73)

GPAI Violations (Art. 51–56)

Failing to draw up Annex XI technical documentation (Art. 53(1)(a))
Failing to provide information to downstream providers (Art. 53(1)(b))
No public copyright compliance policy (Art. 53(1)(c))
No training data summary publication (Art. 53(1)(d))
Systemic risk models: failing to conduct red-teaming (Art. 55(1)(a))
Failing to report serious incidents to EU AI Office (Art. 55(1)(b))
Inadequate cybersecurity measures for systemic risk models (Art. 55(1)(c))

Dimension	GDPR	EU AI Act
Highest fine tier	€20M or 4% of global turnover	€35M or 7% — higher in both absolute and percentage terms
Standard fine tier	€10M or 2% of global turnover	€15M or 3% — again higher in both metrics
Enforcement body	27 national Data Protection Authorities (DPAs), coordinated via EDPB	27 national Market Surveillance Authorities (NMSAs) + EU AI Office (GPAI direct jurisdiction)
Individual rights to sue	Yes — individuals can claim compensation for GDPR violations (Art. 82 GDPR)	No individual private right of action — enforcement is via public authorities only
Enforcement track record	Active since 2018; landmark fines: Meta €1.2B (DPC, 2023), WhatsApp €225M (DPC, 2021)	Active from Aug 2025 (GPAI) / Aug 2026 (high-risk AI). First major fines expected 2027.
Simultaneous application	Both GDPR and the EU AI Act can apply to the same AI system. A single incident can trigger investigations under both frameworks, by different authorities, resulting in separate penalties. Neither law limits the other’s penalty application.

The critical insight: For large technology companies with €100B+ global revenue, a Tier 1 EU AI Act violation yields a maximum fine of €7 billion (7% of €100B). A concurrent GDPR violation at 4% yields €4 billion. Together: up to €11 billion — dwarfing any previous EU regulatory penalty. For mid-size companies with €500M revenue: €35M (AI Act) + €20M (GDPR) = €55M maximum simultaneous exposure. These are theoretical maximums — actual fines will be lower — but they illustrate the regulatory stakes.

5. How Penalties Are Calculated: Mitigating and Aggravating Factors

The figures above are maximums — not automatic fines. Article 99(6) requires national authorities and the EU AI Office to consider the following factors when determining the actual fine amount. Understanding these factors is essential for both calibrating compliance investment and for post-incident response strategy.

⬆ Aggravating Factors (higher fine)

Intentional or serious negligence rather than inadvertent error
Long duration of the violation — ongoing non-compliance
Failure to cooperate with authorities during investigation
Previous violations of the EU AI Act by the same organisation
Large number of affected individuals or high severity of harm
Deliberately obstructing or delaying authority access to information
Continuing the violation after formal notification to stop
Financial benefit derived from the violation

⬇ Mitigating Factors (lower fine)

Voluntary disclosure of the violation to authorities
Immediate corrective action taken upon becoming aware
Full cooperation with the investigation
Technical nature of the violation with no actual harm caused
Existing compliance programme demonstrating genuine effort
SME or startup status (see Section 6)
Novel legal uncertainty in how the Act applies to the specific use case
No prior violations of the EU AI Act or related legislation

The voluntary disclosure advantage

From GDPR enforcement precedent, voluntary disclosure typically results in fines of 30–70% below the maximum applicable for a given violation. For a potential €15M violation, proactive disclosure and cooperation may bring the actual fine to €4.5M–€10.5M. More importantly, voluntary disclosure prevents the investigation costs and reputational damage of a drawn-out regulatory inquiry. Many organisations’ first instinct is to conceal — experienced regulatory counsel consistently advises the opposite.

6. SME and Startup Rules: Reduced Penalties and Proportionality

The EU AI Act explicitly acknowledges that the same penalty structure cannot be applied identically to a €100B multinational and a 10-person startup. Article 99(7) requires national authorities to give particular consideration to the interests and specific needs of SMEs, including startups.

Organisation Type	EU Definition	Penalty Regime
Large enterprise	>250 employees or >€50M annual turnover	Full penalty regime applies. Percentage clause typically dominates for large global operations.
Medium enterprise (SME)	50–250 employees, €10M–€50M turnover	Proportionality applies. Fixed cap dominates. Authorities required to consider financial impact. Generally lower starting points in fine calculations.
Small enterprise / startup	<50 employees, <€10M turnover	Strongest proportionality protections. Authorities must consider viability impact. Corrective action orders often preferred over fines for first-time technical violations.
Micro-enterprise	<10 employees, <€2M turnover	Most lenient treatment. For Tier 2 and 3 violations, corrective actions and compliance orders are the expected default. Fines reserved for persistent non-compliance or serious harm.

Important: SME status does not affect prohibited AI penalties

The proportionality protections for SMEs apply fully to Tier 2 and 3 violations. They do not eliminate Tier 1 fines for prohibited AI practices — these are absolute prohibitions where regulators have very limited discretion to reduce penalties based on company size. A small startup operating facial recognition surveillance in public spaces faces the same fundamental prohibition as Google — though the fine calculation would naturally yield a lower absolute figure due to the lower global turnover base.

7. Beyond Fines: Other Enforcement Powers

Administrative fines are the most financially visible enforcement tool, but they are far from the only one. Article 74 grants national market surveillance authorities a toolkit of enforcement measures that can cause significant operational disruption beyond financial penalties:

🚫

Market withdrawal orders — Authorities can order a non-compliant high-risk AI system removed from the EU market entirely. For B2B SaaS products with EU customers, this is an existential commercial threat that makes fines look minor by comparison. A market withdrawal order for a widely deployed HR AI tool could terminate dozens of customer contracts simultaneously.

🔧

Corrective action orders — Authorities can mandate specific technical or organisational changes to bring a system into compliance — including model modifications, documentation updates, human oversight procedure changes, and access restriction requirements. These can be both costly and operationally disruptive.

📢

Public disclosure orders — Authorities may publicly disclose the violation, the system involved, and the corrective action required. For enterprise software vendors, public disclosure of a non-conformant high-risk AI system can damage customer trust and trigger contract reviews that far outweigh the financial penalty itself.

🔍

Inspection and access rights — Authorities can demand access to AI systems, technical documentation, training data, and internal communications relevant to compliance. Non-compliance with access requests is a Tier 2 violation. The investigation process itself — with its associated legal costs, management time, and operational disruption — is a significant penalty beyond any formal fine.

8. Real-World Penalty Scenarios

These hypothetical scenarios illustrate how the penalty regime applies in practice:

Scenario A: Large HR Tech Vendor — Missing Conformity Assessment

A UK-based HR technology company with €200M annual revenue sells an AI-powered CV screening tool used by 300+ EU employers. The company has not completed a conformity assessment or registered in the EU AI database by August 2026. A German employer’s works council files a complaint with the Bundesnetzagentur (German NMSA).

Violation: Art. 43 (no conformity assessment) + Art. 49 (no EU database registration) — both Tier 2.
Turnover-based maximum: 3% × €200M = €6M per violation.
Likely actual fine: First violation, technically complex area, some compliance effort evident — €1.5M–€3M range, plus a mandatory market withdrawal until compliant. Commercial impact of market withdrawal from 300+ EU deployments likely exceeds the fine significantly.

Scenario B: US AI Company — Real-Time Facial Recognition at EU Retail

A US AI company provides facial recognition surveillance to 50 EU retail chains for “loss prevention.” The company is aware their system constitutes real-time biometric identification in public spaces but has relied on private-sector use as a legal argument. A French civil liberties NGO files a complaint with CNIL’s AI enforcement unit.

Violation: Art. 5(1)(h) — prohibited real-time biometric identification in public spaces — Tier 1.
Global turnover: €2 billion. 7% = €140M maximum.
Likely actual fine: Tier 1 violations carry less mitigating discretion. Intentional deployment of a prohibited practice: €50M–€100M range highly plausible, plus mandatory immediate shutdown. Criminal referrals possible in some member states.

Scenario C: EU Scale-Up — Voluntary Disclosure of Documentation Gap

A German AI startup (€8M revenue) discovers after August 2026 that their credit scoring AI deployed to three EU banks has an incomplete Annex IV file — the bias testing section is missing disaggregated results for two protected characteristics. They immediately notify the German NMSA, suspend the product for affected customer segments, and commission the missing testing.

Violation: Art. 10 + Art. 11 (incomplete data governance documentation) — Tier 2.
Turnover-based maximum: 3% × €8M = €240K. Fixed cap €15M would be far higher — percentage governs for SMEs.
Likely actual fine: Voluntary disclosure, immediate remediation, startup status, no harm caused, technical violation: corrective action order only or nominal fine of €20K–€50K. This is the best-case scenario, achievable through proactive compliance culture.

9. Frequently Asked Questions

Can individuals sue under the EU AI Act to claim compensation? +

No — unlike GDPR’s Article 82 (which gives individuals a private right to claim compensation for data protection violations), the EU AI Act does not include an equivalent individual private right of action. Enforcement is exclusively through public authorities. However, this does not mean individuals have no recourse: they can file complaints with national market surveillance authorities triggering investigations; they retain their GDPR rights for data-related harms from AI; and in some member states, national tort law may create parallel liability for AI-caused harm. The Product Liability Directive revision is also expected to create clearer civil liability pathways for AI-caused harm.

Can we be fined in both GDPR and EU AI Act for the same incident? +

Yes. GDPR and the EU AI Act are separate legal regimes, enforced by different authorities (DPAs vs NMSAs). A single AI incident — for example, a biased HR AI that discriminates against protected groups — could simultaneously be a GDPR violation (Art. 22 automated decision rights) and an EU AI Act violation (Art. 10 inadequate bias detection). Each authority investigates under their own framework and each can impose separate penalties. The EU AI Act does not include a “ne bis in idem” (double jeopardy) clause preventing simultaneous GDPR fines for the same facts. Coordinate your legal response across both frameworks immediately in the event of a serious incident.

Will the EU actually enforce these penalties — or is this like early GDPR with no real action? +

Early GDPR enforcement was slower than anticipated — but it accelerated significantly from 2019 onwards. Total GDPR fines exceeded €4 billion by end of 2025. EU AI Act enforcement is likely to follow a similar trajectory. Several factors suggest it may be faster: the EU AI Office’s centralised GPAI enforcement removes the fragmentation that slowed GDPR; civil society organisations have explicitly announced strategic enforcement complaint plans; and the political salience of AI harms (election disinformation, employment bias, privacy violations) creates pressure for visible enforcement action. Our recommendation: do not plan your compliance programme around enforcement delay. The compliance investment is justified on risk-adjusted grounds regardless of when the first major fine lands. Start with our compliance checklist to understand where you stand today.

Find out your compliance exposure now

The best way to reduce penalty exposure is to complete compliance before enforcement begins. Our risk assessment and checklist get you started in minutes.

Free Risk Assessment → Compliance Checklist →

Sources: EU AI Act Article 99 — EUR-Lex · EU AI Office · GDPR.eu · GDPR Enforcement Tracker · IAPP EU AI Act Resource Centre · Revised Product Liability Directive