CEIAI Governance Database
Back to the database
Record STMT-M033Soft law

Commission Guidelines on the scope of the obligations for providers of general-purpose AI models established by Regulation (EU) 2024/1689 (AI Act)

Guidelines State lawEuropean Unionin force

These guidelines issued by the European Commission clarify the scope of obligations for providers of general-purpose AI (GPAI) models under Regulation (EU) 2024/1689 (AI Act). They detail criteria for classifying GPAI models and models with systemic risk, rules on model lifecycles, and exceptions for open-source releases. Furthermore, the document outlines enforcement mechanisms and compliance expectations following the entry into application of Chapter V of the AI Act on August 2, 2025.

Summary and content fields are drawn from the document text and are pending editorial review.

Who issued it

European CommissionGovernment (executive)
Lead issuer
Legal forceSoft law
Under whose authorityState law
Judged fromfull text

Who it governs

AI developers and providersRegulatorsCompaniesAI deployers and usersResearchers and academics

AI developers, providers, downstream modifiers, and researchers seeking guidance on how to comply with general-purpose AI obligations under the EU AI Act.

Sectors
Cross-sectorTelecommunications and digital services · artificial intelligence

What the document contains

What's covered

The document addresses the definition and classification criteria of general-purpose AI models and models with systemic risk under the EU AI Act. It details the obligations of model providers regarding technical documentation, copyright policies, open-source exceptions, and enforcement mechanisms managed by the European Commission.

What's missing

The guidelines do not apply to other Union laws, such as product liability rules, and do not provide authoritative interpretations reserved exclusively for the Court of Justice of the European Union.

Key takeaways
  • The guidelines set an indicative criterion for a GPAI model based on training compute greater than 10^23 FLOP and specific text, audio, image, or video generation modalities.
  • Models trained with cumulative compute greater than 10^25 FLOP are presumed to present systemic risks and are subject to stringent assessment and mitigation duties.
  • Providers releasing GPAI models under a free and open-source licence are exempt from certain technical documentation obligations, provided the parameters are public and no monetisation strategy is applied.
  • Downstream modifiers who modify an existing model using compute exceeding a third of the original training compute can become providers of a new GPAI model.
  • The European Commission, acting through the AI Office, has exclusive competence to enforce Chapter V obligations starting from August 2, 2025, with fining powers applicable from August 2, 2026.
Outline
· 1. Background and objectives of the Commission guidelines
· 2. General-purpose AI models
  - 2.1. When is a model a general-purpose AI model?
  - 2.2. The lifecycle of a general-purpose AI model
  - 2.3. When is a general-purpose AI model a general-purpose AI model with systemic risk?
· 3. Providers placing on the market general-purpose AI models
  - 3.1. When is an actor a provider placing on the market a general-purpose AI model?
  - 3.2. Downstream modifiers as providers of general-purpose AI models
· 4. Exceptions to certain obligations for certain models released as open-source
  - 4.1. Scope of the exceptions
  - 4.2. Conditions for the exceptions to apply
· 5. Enforcement of the obligations for providers of general-purpose AI models
  - 5.1. Effects of adhering to a code of practice assessed as adequate
  - 5.2. Supervision, investigation, enforcement, and monitoring of Chapter V of the AI Act
  - 5.3. Compliance after entry into application of Chapter V of the AI Act on 2 August 2025
· 6. Review and update of the Commission guidelines
· Annex: Training compute of general-purpose AI models
Issued19 November 2025
Length15,999 words · 36 pages
FormatNot recorded
Scoperegional

How it scores

Pending review

Each score is backed by a quote verified against the document text. Editorial review has not yet confirmed them.

Fairness

0/5

Non-discrimination and equitable treatment

screening score · 2026-10-24

Human Dignity

1/5

Respect for the inherent worth of persons

Its aim is to promote innovation in and uptake of artificial intelligence (‘AI’) in the Union, while ensuring a high level of protection of health, safety, and fundamental rights, including democracy and the rule of law.

Paragraph (1)

screening score · 2026-10-24

Human Oversight

0/5

Humans remain in control of consequential decisions

screening score · 2026-10-24

Privacy

1/5

Protection of personal data and private life

While fully recognising that the protection of personal data is a fundamental right and that therefore personal data cannot be considered a commodity or monetised

Section 4.2.2

screening score · 2026-10-24

Topics

general_purpose_ai ai_act systemic_risk compliance open_source