Join our Newsletter — 33% off our NHI Course
Home› Guides› EU AI Act Penalties and Enforcement Guide: Article…
Guide AI Regulation & Law

EU AI Act Penalties and Enforcement Guide: Article 99 Fines, Market Surveillance Authorities and AI Office Powers

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 18 November 2026 14 min read
On this page

The EU AI Act is enforced on two tracks: national market surveillance authorities supervise AI systems and apply fines set under Article 99, while the European Commission, acting through the AI Office, supervises general-purpose AI models and, after the Digital Omnibus on AI, certain AI systems built on them. This guide sets out the fine tiers, the factors authorities weigh, who enforces what, the procedures they use, and how security and identity teams can reduce exposure, based on the text of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744.

For a programme-level view of the Regulation, start with the EU AI Act Compliance Guide. This page goes deeper on enforcement only.

This guide is not legal advice. Status as of 17 Nov 2026. National penalty laws and authority designations vary by Member State and change over time; confirm with qualified counsel and the relevant authority.

Key takeaways

  • Article 99 sets three ceilings: up to EUR 35,000,000 or 7% of worldwide annual turnover for prohibited practices; a middle tier under Article 99(4) for most operator obligations (confirm the amounts in Article 99(4) on EUR-Lex); up to EUR 7,500,000 or 1% for incorrect, incomplete or misleading information. For undertakings, the higher figure applies.
  • For SMEs and start-ups the lower figure applies (Article 99(6)), and the Digital Omnibus extended the same rule to small mid-cap enterprises for the Article 99(4) and 99(5) tiers (new Article 99(6a)).
  • Member States lay down the penalty rules and decide who imposes fines; fines must be effective, proportionate and dissuasive (Article 99(1) and (9)).
  • The Commission can fine general-purpose AI model providers up to EUR 15,000,000 or 3% (Article 101), and the European Data Protection Supervisor can fine Union institutions (Article 100).
  • Article 99 applied from 2 August 2025 under Article 113(b). Most high-risk obligations now apply from 2 December 2027 (Annex III) and 2 August 2028 (Annex I), so the high-risk exposure grows from those dates.
  • Cooperation, self-notification and the technical and organisational measures in place are explicit fine factors (Article 99(7)). Good identity and logging controls are mitigation evidence.

Enforcement at a glance

WhoSupervisesMain powersLegal basis
National market surveillance authoritiesAI systems, including high-risk and prohibited practicesEvaluation, corrective action, withdrawal, recall, fines under national lawArticles 70, 74, 79 to 83, 99
Sectoral authoritiesAnnex I products under their product law; regulated financial institutionsMarket surveillance for their sectorsArticle 74(3) and (6)
Data protection authoritiesSpecified Annex III systems, including law enforcement, border management and justice usesMarket surveillance for those systemsArticle 74(8)
European Commission and AI OfficeGeneral-purpose AI models; after the Omnibus, certain AI systems built on themInformation requests, evaluations, measures, finesArticles 75, 88 to 94, 101
European Data Protection SupervisorUnion institutions, bodies, offices and agenciesMarket surveillance and finesArticles 74(9), 100
European Artificial Intelligence BoardConsistency across Member StatesAdvice and coordination; no fining powerArticle 65

Scope and timing

Penalties apply to operators: providers, deployers, importers, distributors, authorised representatives and product manufacturers, plus notified bodies for their own obligations. Chapter XII, which contains Articles 99 to 101, applied from 2 August 2025 under Article 113(b), except Article 101, so national penalty regimes for prohibited practices and for the GPAI-related and governance provisions already in force could be used from that date. The Digital Omnibus added new point (d) to Article 113 so that Articles 102 to 110 apply from 27 July 2026.

The high-risk obligations in Chapter III Sections 1 to 3 apply from 2 December 2027 for Annex III systems and from 2 August 2028 for Annex I systems under Article 113(c) as amended. New prohibitions added to Article 5(1) by the Omnibus, covering non-consensual intimate imagery and child sexual abuse material, apply from 2 December 2026. The EU AI Act Implementation Timeline Tracker tracks each date.

Article 99 fine tiers

Tier 1: prohibited practices

Non-compliance with the prohibition of AI practices in Article 5 is subject to administrative fines of up to EUR 35,000,000 or, for an undertaking, up to 7% of total worldwide annual turnover for the preceding financial year, whichever is higher (Article 99(3)). The EU AI Act Prohibited Practices Guide describes the practices.

Tier 2: operator and notified body obligations

Article 99(4) sets a middle tier of fines (confirm the amounts in the EUR-Lex text) for non-compliance with listed provisions other than Article 5. The list covers provider obligations (Article 16), authorised representatives (Article 22), importers (Article 23), distributors (Article 24), deployers (Article 26), notified body requirements and obligations (Articles 31, 33(1), (3) and (4), and 34) and transparency obligations for providers and deployers (Article 50). The Digital Omnibus added point (da), covering obligations of providers and operators under Article 25(2) and (4) on the value chain.

Tier 3: information to authorities

Supplying incorrect, incomplete or misleading information to notified bodies or national competent authorities in reply to a request is subject to fines of up to EUR 7,500,000 or 1%, whichever is higher (Article 99(5)).

SMEs and small mid-caps

For SMEs, including start-ups, each fine is up to the percentage or amount, whichever is lower (Article 99(6)). The Digital Omnibus defined small mid-cap enterprises in new Article 3(14b) and added Article 99(6a), applying the lower-of rule to them for the fines in paragraphs 4 and 5. The prohibited-practice tier keeps the higher-of rule for small mid-caps.

InfringementCapLarge undertakingSME and start-upSmall mid-cap
Article 5 prohibited practiceEUR 35m or 7%HigherLowerHigher
Article 99(4) obligations, including Article 26Middle tier (see Article 99(4) on EUR-Lex)HigherLowerLower
Incorrect or misleading informationEUR 7.5m or 1%HigherLowerLower

How fines are set: the Article 99(7) factors

When deciding whether to impose a fine and its amount, authorities take into account all relevant circumstances of the specific situation and, as appropriate, regard to:

  • the nature, gravity and duration of the infringement and its consequences, taking into account the purpose of the AI system and the number of affected persons and level of damage;
  • whether other market surveillance authorities have already fined the same operator for the same infringement;
  • whether other authorities have fined the same operator for infringements of other Union or national law arising from the same activity;
  • the size, annual turnover and market share of the operator;
  • other aggravating or mitigating factors, such as financial benefits gained or losses avoided;
  • the degree of cooperation with national competent authorities to remedy the infringement and mitigate its effects;
  • the degree of responsibility, taking into account the technical and organisational measures implemented;
  • the manner in which the infringement became known, in particular whether and to what extent the operator notified it;
  • the intentional or negligent character of the infringement;
  • any action taken to mitigate the harm suffered by affected persons.

Several factors reward preparation: documented technical and organisational measures, early self-notification and fast cooperation. The Digital Omnibus also replaced Article 99(1) and its recitals stress respect for the ne bis in idem principle where the same conduct is criminally sanctioned.

National enforcement: who your regulator is

Article 70 requires each Member State to establish or designate at least one notifying authority and at least one market surveillance authority as national competent authorities. Member States were to make contact information publicly available by 2 August 2025 and report on authority resources by that date and every two years after. In practice this means the competent authority differs by country and, within a country, by sector.

Article 74 shapes the allocation:

  • Regulation (EU) 2019/1020 on market surveillance applies to AI systems, so familiar product-safety tools apply.
  • For high-risk systems related to Annex I Section A products, the authority responsible under that product legislation is generally the market surveillance authority (Article 74(3)).
  • For high-risk systems placed on the market, put into service or used by financial institutions regulated by Union financial services law, the financial supervisors act as market surveillance authorities (Article 74(6)).
  • For specified Annex III systems used for law enforcement, border management, justice and democratic processes, Member States designate data protection authorities or other authorities under the same conditions (Article 74(8)).
  • The European Data Protection Supervisor is the market surveillance authority for Union institutions (Article 74(9)).

Article 99(9) leaves it to national law whether fines are imposed by courts or other bodies, and Article 99(8) lets each Member State decide whether and to what extent fines apply to its own public authorities. Check both points in the national implementing law for each country where you operate. Member States must notify their penalty rules to the Commission and report annually on fines imposed (Article 99(2) and (11)).

Enforcement procedures for AI systems

Systems presenting a risk: Article 79

Where a market surveillance authority has sufficient reason to consider that an AI system presents a risk to health, safety or fundamental rights, it evaluates the system. If it finds non-compliance it requires the operator to take corrective action, withdraw or recall the system within a period it prescribes, and in any event within 15 working days or as provided in relevant Union harmonisation law. If the operator does not act, the authority can restrict, prohibit, withdraw or recall the system. Other Member States and the Commission can object within three months, or 30 days for Article 5 infringements.

Misclassified non-high-risk systems: Article 80

Authorities can evaluate a system that its provider classified as non-high-risk under Article 6(3). If the system is high-risk, the provider must bring it into compliance, and fines under Article 99 apply where the misclassification was to circumvent the requirements (Article 80(7)). See the EU AI Act Article 6(3) Guide.

Compliant systems that still present a risk: Article 82

Even a compliant high-risk system can be subject to measures if it presents a risk to health, safety or fundamental rights. The operator may be required to remove the risk or withdraw or recall the system, and the Commission evaluates whether the national measure is justified.

Complaints and explanations: Articles 85 and 86

Any natural or legal person with grounds to consider an infringement has occurred can submit a complaint to the relevant market surveillance authority (Article 85). Affected persons have a right to an explanation from the deployer of certain Annex III high-risk decisions (Article 86). Complaints are a likely trigger for investigations, so complaint handling and explanation processes are enforcement controls.

The AI Office and Commission powers

Under Article 88, the Commission has exclusive powers to supervise and enforce Chapter V on general-purpose AI models, and entrusts those tasks to the AI Office. Its toolkit includes:

  • Monitoring of compliance, including through approved codes of practice; downstream providers can lodge reasoned complaints (Article 89).
  • Qualified alerts from the scientific panel on models that pose concrete risks at Union level or meet systemic-risk conditions (Article 90, with the panel set up under Article 68).
  • Requests for documentation and information (Article 91).
  • Evaluations of models, including through APIs or other technical means and access to source code, with independent experts (Article 92).
  • Measures requiring compliance, risk mitigation, or restriction, withdrawal or recall of a model, or acceptance of binding commitments (Article 93).
  • Procedural rights for providers through Article 18 of Regulation (EU) 2019/1020 (Article 94).

Article 101 lets the Commission fine providers of general-purpose AI models up to 3% of annual total worldwide turnover or EUR 15,000,000, whichever is higher, where it finds the provider intentionally or negligently infringed the Regulation, failed to comply with a request under Article 91 or supplied incorrect, incomplete or misleading information, failed to comply with a measure under Article 93, or failed to give access for an evaluation under Article 92. The Commission must communicate preliminary findings and give the provider an opportunity to be heard, and the Court of Justice has unlimited jurisdiction to cancel, reduce or increase the fine. The EU AI Act GPAI Obligations Guide covers the underlying duties.

The Digital Omnibus extended the AI Office's role. Amended Article 75(1) makes it exclusively competent for certain AI systems, including systems based on a general-purpose AI model where model and system come from the same provider, and AI systems integrated into designated very large online platforms or search engines, subject to listed exceptions. New paragraphs 1a to 1e cover serious incident reporting to the AI Office, assistance from national authorities, and AI Office conformity assessments for some high-risk systems. If you build agents or applications on your own foundation model, expect the AI Office rather than a national authority to be your supervisor for those systems.

Union institutions: Article 100

The European Data Protection Supervisor can impose fines on Union institutions, bodies, offices and agencies: up to EUR 1,500,000 for Article 5 infringements and up to EUR 750,000 for other non-compliance. The institution must be heard first, and fines go to the general budget of the Union without affecting the institution's effective operation.

What it means for security and identity teams

Several of the obligations that attract Tier 2 fines are operational controls security and identity teams already own, and the Article 99(7) factors reward evidence of them:

  • Logs and record keeping. Provider logging (Article 12) and deployer log retention (Article 26(6)) let you show what happened, who or what acted, and when. Logs that attribute actions to human and non-human identities support both incident reconstruction and a credible cooperation story.
  • Kill switches. Article 79 corrective action windows can be short. Being able to revoke an AI system's credentials or suspend an agent quickly makes withdrawal practical.
  • Accurate information. The 1% tier punishes misleading answers. Keep an authoritative inventory of AI systems, their identities and data access so replies to authorities are correct the first time.
  • Incident notification. Self-notification is a mitigating factor. Integrate AI serious incidents into incident response with clear owners.

Deployer-specific duties are covered in the EU AI Act Article 26 Deployer Obligations Guide.

Common failure patterns

  • Assuming the EUR 35m or 7% ceiling is the only number that matters, while the more likely exposure is the Article 99(4) tier for operational obligations.
  • Not knowing which national authority supervises each system in each country, especially where financial or data protection authorities take the role.
  • Treating information requests as routine correspondence and answering from memory, risking the 1% tier.
  • No way to withdraw or suspend a system within an authority's corrective action period.
  • Building systems on an in-house foundation model without realising AI Office supervision may apply after the Digital Omnibus.
  • Ignoring complaint and explanation processes, which are routes by which authorities learn of issues.

Practitioner checklist

  • Map each AI system and model to its Article 99 tier exposure and the date that exposure begins.
  • Identify the competent authority for each system per Member State, including sectoral, financial and data protection authorities.
  • Check each national implementing law for who imposes fines and whether public authorities can be fined.
  • Determine whether you qualify as an SME or small mid-cap and record the evidence.
  • Establish a single owner and process for responding to authority information requests, with legal review.
  • Test the ability to suspend, withdraw or recall each system, including credential revocation for its identities.
  • Integrate AI serious incidents and self-notification into incident response.
  • Keep evidence of technical and organisational measures for use as Article 99(7) mitigation.
  • Check whether any of your systems fall within AI Office competence under amended Article 75.

Standards and references

Related NHI Mgmt Group resources: EU AI Act Compliance Guide · EU AI Act Executive Briefing · EU AI Act High-Risk AI Systems Guide · EU AI Act Readiness Checklist

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 18 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.