Docs

Appearance

EU AI Act — Commission guidance ​

The European Commission gives guidance on Regulation (EU) 2024/1689, the EU AI Act. This section shows selected Commission guidance. The guidance is on five areas: the definition of an AI system, prohibited AI practices, high-risk classification, transparency, and general-purpose AI (GPAI) models. It also gives the publication and adoption of each guidance document, the application dates of the EU AI Act, and its enforcement and fines.

Use the guidance ​

Each page of this section starts from the EU AI Act and adds the Commission guidance for one decision:

  • Is the system an AI system? See Definition of an AI system for the Article 3(1) definition and the Article 2 scope rules.
  • Is the practice prohibited? The Commission guidelines of 2025 give the Commission interpretation of Article 5(1), points (a) to (h), of the 2024 text. Regulation (EU) 2026/1744 added points (ba) and (bb) and Article 5(1a) and (1b), which apply from 2 December 2026. See Prohibited AI practices.
  • Is the AI system high-risk? See High-risk classification for Article 6(1) and (2) and the Article 6(3) exception for Annex III systems.
  • Find Commission examples for your intended use. The draft examples for Annex I and Annex III apply only together with the conditions of each Annex point. See High-risk examples.
  • Which transparency duty applies? Article 50(1) and (2) give duties to providers. Article 50(3) and (4) give duties to deployers. See Transparency obligations.
  • Is the model a GPAI model? See GPAI models for the Article 3(63) definition.

For the provisions of the EU AI Act, see EU AI Act.

Commission guidance ​

The European Commission gives its guidance on the EU AI Act in guidelines and other documents. The EU AI Act, Regulation (EU) 2024/1689, entered into force on 1 August 2024. Regulation (EU) 2026/1744 of 8 July 2026, the Digital Omnibus on AI, amended the EU AI Act. Regulation (EU) 2026/1744 is in the Official Journal of 24 July 2026 and entered into force on 27 July 2026 (Article 4). The consolidated text of the EU AI Act of 27 July 2026 contains these amendments.

Article 96 and Article 6(5) ​

Article 96(1) of the EU AI Act tells the Commission to write guidelines on the "practical implementation" of the Act. Article 96(1) names areas for these guidelines, for example:

  • point (b): the prohibited practices of Article 5
  • point (d): the transparency obligations of Article 50
  • point (f): the definition of an AI system in Article 3(1).

Article 6(5) tells the Commission to give guidelines on high-risk classification, with "practical examples of use cases of AI systems that are high-risk and not high-risk".

For the Article 96(1) guidelines, the Commission must "involve the Board". The Commission must give the Article 6(5) guidelines "after consulting the European Artificial Intelligence Board". Article 96(2) tells the Commission to "update guidelines previously adopted when deemed necessary". The Commission does this "At the request of the Member States or the AI Office, or on its own initiative".

The guidance is not binding ​

Commission guidelines on the EU AI Act are not binding. Paragraph (7) of the Commission guidelines on the definition of an AI system gives this rule:

The Guidelines are not binding. Any authoritative interpretation of the AI Act may ultimately only be given by the Court of Justice of the European Union (CJEU).

Paragraph (5) of the guidelines on prohibited practices and paragraph (9) of the guidelines on GPAI models contain the same rule. The guidelines on prohibited practices are for the "consistent, effective and uniform application" of Article 5 (paragraph (4)). This does not make them binding in a Member State.

The EU AI Act, as amended by Regulation (EU) 2026/1744, is the law. Commission guidance does not change it. The definition guidelines, prohibited-practices guidelines, and GPAI-scope Guidelines are from 2025, before the 2026 amendments. Thus, read them together with the amended text. For example, paragraph (96) of the GPAI-scope Guidelines refers to Article 56(6) as it was before Regulation (EU) 2026/1744 replaced that paragraph.

The GPAI-scope Guidelines also tell providers that "a case-by-case assessment will always be necessary" (paragraph (9)). Modulos recommends that you record your "assessment", how you made it, and where you do not follow the Commission interpretation. This record is a Modulos recommendation. It is not a duty of the EU AI Act. This record does not show that you obey the Act. A draft guidance document can change before its adoption by the Commission.

Publication and adoption dates ​

These are the publication and adoption dates of the Commission guidance on the EU AI Act that these pages show:

GuidancePublicationAdoption by the CommissionType
Commission Guidelines on the definition of an AI systemOn 6 February 2025, with content approval in Communication to the Commission "C(2025) 924 final"On 29 July 2025, Communication "C(2025) 5053 final"Adopted
Commission Guidelines on prohibited AI practicesOn 4 February 2025, with content approval in Communication to the Commission "C(2025) 884 final"On 29 July 2025, Communication "C(2025) 5052 final"Adopted. For Article 5(1), points (a) to (h), of the 2024 text
Commission Guidelines on the scope of the obligations for providers of GPAI modelsOn 18 July 2025, with content approval in Communication to the Commission "C(2025) 5045 final"On 19 November 2025, Communication "C(2025) 7719 final"Adopted
Draft Commission guidelines on the classification of high-risk AI systemsOn 19 May 2026, in three sectionsThe Commission plans adoption by the end of 2026Draft for consultation
Guidelines on the transparency obligations of Article 50On 20 July 2026Content approved on 20 July 2026, Communication to the Commission C(2026) 5054 finalSee the condition below

The Commission extended the high-risk draft consultation from 23 June 2026 to 23 July 2026.

The Commission published the draft in three documents, one for each section: "General principles", "Annex I of AI Act", and "Annex III of AI Act". The Commission tells you that, because of this, "you do not need to consult the full document" for one area.

On 20 July 2026, the Commission approved the content of its Article 50 guidelines in Communication to the Commission C(2026) 5054 final. The guidelines are the Annex to that Communication. The Communication gives this condition:

The Guidelines will be formally adopted by the Commission at a later date, when all language versions are available. It is only from that moment that these Guidelines will be applicable.

The approval of 20 July 2026 does not show if the Commission adopted the guidelines after that date. The Commission publication of 20 July 2026 for these guidelines tells you that "The Commission adopted these guidelines". The Commission document IP/26/1714 of 31 July 2026 shows them as "Draft guidelines on the transparency obligations for certain AI systems". Article 50 of the EU AI Act applies from 2 August 2026, with the Article 111(4) date for Article 50(2) (see Application dates).

Other Commission guidance ​

These Commission guidance pages do not show all Commission guidance on the EU AI Act. Other examples are:

Article 96(1)(g) tells the Commission to give guidelines on Article 8(2), Article 9(10), and Article 17(3) for the "Union harmonisation legislation listed in Section A of Annex I". The Act tells the Commission that "such guidelines shall be published by 1 August 2027".

The Commission also names more guidelines that the AI Office prepares, for example on "responsibilities along the AI value chain" and on the "provisions related to substantial modification". These guidelines do not change the provisions of the Act. For example, Article 3(23) gives the definition of substantial modification, and Article 25 gives the rules on "Responsibilities along the AI value chain".

Application dates ​

Different provisions of the EU AI Act apply from different dates (Article 113). The dates of the Act are different from the dates of the Commission guidance. These are the dates of the EU AI Act as amended by Regulation (EU) 2026/1744:

DateProvisions that applyConditions and provision
From 2 February 2025The provisions of Chapter I and Chapter II, for example the Article 3(1) definition and Article 5(1), points (a) to (h)Not Article 5(1), points (ba) and (bb), and not Article 5(1a) and (1b) (Article 113, point (a))
From 2 August 2025Chapter V on GPAI models, Chapter III Section 4, Chapter VII, Chapter XII, and Article 78Not Article 101 (Article 113, point (b)). For GPAI models placed on the market before 2 August 2025, see 2 August 2027
From 27 July 2026Articles 102 to 110Article 113, point (d)
From 2 August 2026All other provisions. Examples are Articles 6(5), 49, 50, 60a, 101, 111, and 112, and the Commission powers in Articles 91 to 93Article 113. For Article 50(2), see Article 111(4)
From 2 December 2026Article 5(1), points (ba) and (bb), and Article 5(1a) and (1b)Article 113, point (a)
By 2 December 2026Article 50(2), for providers of AI systems, also general-purpose AI systems, "generating synthetic audio, image, video or text content" that they placed on the market before 2 August 2026Article 111(4)
By 2 August 2027Chapter V obligations, for providers of GPAI models placed on the market before 2 August 2025Article 111(3)
From 2 December 2027Chapter III Sections 1, 2, and 3, but not Article 6(5), for AI systems classified as high-risk in Article 6(2) and Annex IIIArticle 113, point (c)
From 2 August 2028Chapter III Sections 1, 2, and 3, but not Article 6(5), for AI systems classified as high-risk in Article 6(1) and Annex IArticle 113, point (c). For Section B of Annex I, Article 2(2) limits the provisions that apply
By 2 August 2030Providers and deployers of pre-existing high-risk AI systems intended for use by public authoritiesArticle 111(2): systems placed on the market or put into service before the applicable Chapter III date, other than Article 111(1) systems. The rules can apply earlier after "significant changes in their designs" from the applicable Chapter III date, as set out below
By 31 December 2030AI systems that are components of the "large-scale IT systems" in Annex X, placed on the market or put into service before 2 August 2027Article 111(1), "without prejudice to the application of Article 5"

The EU AI Act gives these conditions for the dates:

  • Article 111(4) applies to all of Article 50(2), also for interactive AI systems that make synthetic content. Article 50(2) tells providers to make sure that outputs are "marked in a machine-readable format and detectable as artificially generated or manipulated". Article 111(4) does not give more time for the Article 50(1) duties, which apply from 2 August 2026.
  • Article 111(2) applies to high-risk AI systems placed on the market or put into service before the Chapter III date of Article 113. It does not apply to the Annex X systems of Article 111(1). The Act applies to the operators of these systems only if, from that date, the systems "are subject to significant changes in their designs". This rule is "without prejudice to the application of Article 5". Providers and deployers of high-risk AI systems intended for use by public authorities must comply by 2 August 2030.
  • Article 49 applies from 2 August 2026. Article 49(2) applies to an AI system "for which the provider has concluded that it is not high-risk according to Article 6(3)". Article 6(3) applies to Annex III systems from 2 December 2027. Annex VIII Section B gives the information for the Article 49(2) registration. Regulation (EU) 2026/1744 removed its points 7 and 9.
  • Section B of Annex I: Article 2(2) limits which provisions apply to Article 6(1) high-risk AI systems related to Annex I Section B products.
  • Chapter V: the Chapter III dates of 2027 and 2028 do not change the Chapter V dates. They also do not change the 2 August 2026 date of the Commission powers for GPAI models.

Enforcement and penalties ​

Different authorities enforce the EU AI Act. The Commission publication "The enforcement framework of the AI Act" names three: the AI Office of the Commission, the European Data Protection Supervisor (EDPS), and the national competent authorities.

  • National competent authorities enforce the EU AI Act for AI systems that are not in the competence of the AI Office or the EDPS.

  • The AI Office is exclusively competent for the supervision and enforcement of the EU AI Act for two groups of AI systems (Article 75(1)):

    • AI systems "based on general-purpose AI models where the model and the system are developed by the same provider, or by providers forming part of the same undertaking".

      Article 75(1)(a) gives four exceptions:

      • AI systems related to products in Annex I
      • AI systems in Annex III point 2
      • AI systems "provided by" law enforcement authorities, "border management authorities", and financial institutions, "insofar as those AI systems fall under Article 74(6)"
      • AI systems in Annex III point 8 "as regards the administration of justice".

    • AI systems that are, or are "integrated into", a "very large online platform or very large online search engine designated in accordance with Regulation (EU) 2022/2065".

    This competence applies to the providers of these systems. It applies to their deployers only when the deployer is also the provider or is part of the same undertaking as the provider.

  • The Commission has "exclusive powers to supervise and enforce Chapter V" for providers of GPAI models. The Commission entrusts these tasks to the AI Office (Article 88(1)).

  • The EDPS enforces the EU AI Act for AI systems of the "Union institutions, bodies, offices and agencies", and can impose fines on them (Article 100(1)).

The Commission guidance has these functions in enforcement. Paragraph (9) of the GPAI-scope Guidelines gives the Commission interpretation "on which it will base its enforcement action". The Commission publication on the high-risk guidelines tells you that the draft guidelines "will guide enforcement". Article 99(1) tells the Member States to make and apply their rules on penalties "taking into account the guidelines issued by the Commission pursuant to Article 96".

Fines ​

The Member States make the rules on penalties for infringements of the EU AI Act by operators (Article 99(1)). The EU AI Act gives these maximum fines:

InfringementMaximum fineProvision
Prohibited AI practices (Article 5)EUR 35 000 000 or, for an undertaking, 7% of its total worldwide annual turnover for the preceding financial year, "whichever is higher". For small and medium-sized enterprises (SMEs), also start-ups, the lower of the two appliesArticle 99(3) and (6)
Transparency obligations of providers and deployers (Article 50)EUR 15 000 000 or, for an undertaking, 3% of its total worldwide annual turnover for the preceding financial year, "whichever is higher". For SMEs, also start-ups, and for small mid-cap enterprises (SMCs), the lower of the two appliesArticle 99(4)(g), (6), and (6a)
"Union institutions, bodies, offices and agencies" (fines by the EDPS)EUR 1 500 000 for Article 5. EUR 750 000 for other requirements and obligations, for example Article 50Article 100(2) and (3)
Providers of GPAI models (fines by the Commission)3% of the annual total worldwide turnover in the preceding financial year or EUR 15 000 000, "whichever is higher"Article 101(1)

Article 101(1) applies when the Commission finds that the provider of a GPAI model "intentionally or negligently" did one of these:

  • infringed "the relevant provisions" of the Act
  • did not obey a "request" for a document or for information in Article 91, or gave "incorrect, incomplete or misleading information"
  • did not obey a measure in Article 93
  • did not give the Commission access to the model for an evaluation in Article 92.

GPAI model enforcement from 2 August 2026 ​

The Chapter V obligations of the EU AI Act for providers of GPAI models apply from 2 August 2025 (Article 113, point (b)). Providers of GPAI models placed on the market before 2 August 2025 must comply by 2 August 2027 (Article 111(3)). From 2 August 2026, the Commission can use these powers:

  • "request" the documentation and information of the provider (Article 91)
  • "request" measures "where necessary and appropriate", for example to "restrict the making available on the market, withdraw or recall the model" (Article 93)
  • impose fines (Article 101).

The AI Office can do evaluations of a GPAI model "after consulting the Board" for one of these purposes (Article 92(1)):

  • to "assess compliance" of the provider with obligations of the Act, where the information gathered in Article 91 "is insufficient"
  • for an investigation of systemic risks at Union level of GPAI models with systemic risk, "in particular following a qualified alert from the scientific panel" in Article 90(1)(a).

Paragraph (112) of the GPAI-scope Guidelines tells providers that before 2 August 2026, the Commission "cannot take any enforcement actions". The same paragraph tells providers that this "lack of enforcement does not put into question the applicability of the obligations" of Chapter V. In paragraph (109), the Commission "invites providers to immediately and proactively contact the AI Office".

Use Modulos ​

Modulos gives the EU AI Act as two Frameworks: OFF-1 at the Organization level and MFF-1 at the AI System level. For the decisions, the steps, and the Requirements of each Framework, see Operationalizing the EU AI Act in Modulos.

FAQ ​

For the FAQ about the Commission guidance on the EU AI Act, see EU AI Act Commission guidance FAQ.

Sources ​

Disclaimer

This page gives general information. It is not legal advice.