European UnionAI Act, Article 53 (obligations for providers of general-purpose AI models)
Regulation (EU) 2024/1689, Article 53
In force since , under challenge in court.
An AI training data rule binding public and private bodies.
- How the hook was established
- express
- What makes it apply
- Market targeting
- Obligation class
- Disclosure, Governance, TDM
- Audit expectation
- on_request
- Who audits it
- Regulator
- Court action
A suit challenging this law is pending. No court has blocked it. Premium data
What it requires
- Publish a training-content summary if you provide a general-purpose modelNIST AI RMFMAP 2.3
- Adopt and follow a copyright policy that respects text and data mining (TDM) opt-outsNIST AI RMFMAP 4.1
- Keep technical documentation available for the AI Office, unless your model is released under a free and open-source licence that allows access, usage, modification and distribution, has its parameters, including the weights, and the information on its architecture and usage made publicly available, and does not carry systemic riskNIST AI RMFGOVERN 1.1
- Draw up, keep up to date and make available to providers of AI systems who intend to integrate your general-purpose AI model into their AI systems information and documentation that enables them to have a good understanding of the model's capabilities and limitations and to comply with their own obligations under the Regulation, and that contains at least the elements set out in Annex XII: a general description of the model (its intended tasks and the type and nature of AI systems it can be integrated into, the acceptable use policies, the date of release and methods of distribution, how it interacts with hardware or software that is not part of the model, the versions of relevant software, its architecture and number of parameters, the modality and format of its inputs and outputs, and its licence) and a description of its elements and development process (the technical means required to integrate it, the modality, format and maximum size of its inputs and outputs, and information on the data used for training, testing and validation). This duty is without prejudice to the need to observe and protect intellectual property rights and confidential business information or trade secrets under Union and national law, and it does not apply to a model released under a free and open-source licence that allows access, usage, modification and distribution and whose parameters, including the weights, the information on the model architecture and the information on model usage are made publicly available, unless the model carries systemic risk (Article 53(1), point (b), and Article 53(2)).
- Make the technical documentation you keep contain at least the information set out in Annex XI, as appropriate to the size and risk profile of the model: a general description of the model (its intended tasks and the type and nature of AI systems it can be integrated into, the acceptable use policies, the date of release and methods of distribution, its architecture and number of parameters, the modality and format of its inputs and outputs, and its licence); a detailed description of its elements and development process (the technical means required to integrate it, the design specifications and training process, information on the data used for training, testing and validation, the computational resources and training time used to train it, and the known or estimated energy consumption of the model, which may be based on information about the computational resources used where the energy consumption is unknown); and, if the model carries systemic risk, the additional Annex XI Section 2 information (a detailed description of the evaluation strategies and results, of the measures put in place for internal or external adversarial testing and of model adaptations, and, where applicable, of the system architecture).
- Cooperate as necessary with the Commission and the national competent authorities in the exercise of their competences and powers under the Regulation (Article 53(3)).
- Apply the Article 53 duties from , when Chapter V starts to apply (Article 113, point (b)). If your general-purpose AI model was placed on the market before , take the necessary steps to comply with the obligations laid down in the Regulation by (Article 111(3)).
These lines are LexLint's reading of the law's text. Also read by: Commission Guidelines on the scope of the obligations for providers of general-purpose AI models under the AI Act (guidance from the European Commission, issued ); European Commission (guidance, recorded ).
In the frameworks
- NIST AI RMF
- GOVERN 1.1Legal and regulatory requirements involving AI are understood, managed, and documented.
- MAP 2.3Scientific integrity and TEVV considerations are identified and documented, including those related to experimental design, data collection and selection (e.g., availability, representativeness, suitability), system trustworthiness, and construct validation.
- MAP 4.1Approaches for mapping AI technology and legal risks of its components – including the use of third-party data or software – are in place, followed, and documented, as are risks of infringement of a third party’s intellectual property or other rights.
- NIST AI 600-1
- GAI-RISK-07Human-AI Configuration
- GAI-RISK-10Intellectual Property
- GAI-RISK-12Value Chain and Component Integration
- NIST AI 200-2 Draft
- OWASP LLM Top 10
- LLM03:2026Excessive Agency
- OWASP Agentic Top 10
- ASI09Human-Agent Trust Exploitation
- MIT mitigations Draft
- FINOS AIGF
- AIR-PREV-005System Acceptance Testing
- AIR-PREV-006Data Quality & Classification/Sensitivity
- AIR-PREV-007Legal and Contractual Frameworks for AI Systems
- AIR-PREV-024Human-in-the-Loop Action Approval Gate
- AIR-DET-004AI System Observability
- AIR-DET-011Human Feedback Loop for AI Systems
1 more
- AIR-DET-021Agent Decision Audit and Explainability
- ETSI EN 304 223
- 5.1.4-1Capability for human oversight designed in and maintained.
- 5.2.1-1Full inventory of assets, including their dependencies and connections.
- 5.2.4-1Developers keep an audit trail of system design and post-deployment maintenance plans and offer it downstream.
- 5.2.4-2.1Minimum record for training data: its source, such as a page URL, and when it was collected.
- 5.4.2-1Operators log system and user actions.
Our reading of where this law's duties sit in each framework, and in the NIST AI RMF, line by line beside each duty above. It never says that running a control meets the law. How we read a framework against the law
Identifiers refer to ETSI EN 304 223 V2.1.1 (2025-12), © ETSI 2025, all rights reserved. ETSI owns the standard; no ETSI text (titles or sentences) is reproduced here. The standard is free to download from ETSI.
If you get it wrong
Criminal exposureNo
Private right of actionNo
Penalty structure
Article 101(1): fines imposed by the Commission on providers of general-purpose AI models; applies from , the Regulation's general date of application, since Article 113, point (b), excepts Article 101 from the earlier date it gives Chapter V.
- Rule
- Higher of
- As of
- Currency
- EUR
- Fixed cap
- 15,000,000
- Turnover percentage cap
- 3
Who enforces it
Enforcement body
European Commission (AI Office)
Enforcement record
The AI Office's own overview page names its Article 101 fining power over general-purpose AI model providers, and its Article 91 to 93 powers to request information, evaluate a model and require corrective measures, as authority it holds; that supervisory and enforcement authority became operative on , the Regulation's general date of application, since Article 113, point (b), excepts Article 101 from the earlier date it gives Chapter V and no earlier date is given to Chapter IX, a year after the Article 53 duties themselves took effect. The same page's own 'Latest News' list, current to , names no request for information, model evaluation, corrective-measures order, or fine issued against a general-purpose AI model provider under Chapter V in the weeks since. The AI Office's own page shows no action under Chapter V over the period from to , which does not mean none was sought.
- As of
- Source link
- https://digital-strategy.ec.europa.eu/en/policies/ai-office
Settledness
Case C-829/26 (Mobilguru v Google Ireland), a reference from the Fővárosi Törvényszék lodged on , refers questions on Article 53(1)(c) of this Regulation, in relation to the training and functioning of chatbots based on artificial intelligence. The Court of Justice lists the case as pending.
- As of
- Guidance link
- https://digital-strategy.ec.europa.eu/en/faqs/guidelines-obligations-general-purpose-ai-providers
- Guidance body
- European Commission
- Under challenge
- Yes
- Open questions
- Does a model trained with more than 10^23 floating point operations that can generate language, text-to-image or text-to-video, but lacks significant generality, count as a 'general-purpose AI model' under Article 3, when the Commission calls that criterion indicative and says a case-by-case assessment will always be necessary?
- Is a modifier or fine-tuner that uses no more than one-third of the original model's training compute outside the Article 53 provider duties, when that threshold is stated in the Commission's guidelines, which the Commission says are not legally binding?
- Does the copyright policy that Article 53(1)(c) requires a provider to put in place reach the use of press publications in the functioning of a chatbot, and not only in its training, the subject of the questions referred in Case C-829/26?
What it makes you log
Who may demand the log
Regulator, Counterparty
What the log must hold
System identity, Data lineage
Logging duty
Article 53 requires a provider of a general-purpose AI model to draw up and keep up to date technical documentation for the AI Office and national authorities, share technical information with providers who integrate the model downstream, and publish a summary of the content used to train the model. The Article never uses the words logs, records, or audit trail, but none of the three duties can be met without maintaining documentation and producing it on request or publishing it.
- Kind
- Implicit
- As of
- Provision
- Article 53(1), points (a), (b) and (d)
Why the legislator wanted it
Recital 107 states the purpose of the training-content summary in Article 53(1)(d). The technical documentation in Article 53(1)(a) has a different stated purpose: Article 53(1)(a) provides it for the AI Office and the national competent authorities upon request, and Recital 101 adds that the information given to downstream providers is meant to give them a good understanding of the model, to enable its integration and to fulfil their own obligations.
- As of
- Quote
- this summary should be generally comprehensive in its scope instead of technically detailed to facilitate parties with legitimate interests, including copyright holders, to exercise and enforce their rights under Union law
- Source link
- https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
- Source kind
- Recital
- Purpose
- individual_redress
- provenance_and_authenticity
Also on the record
EEA status
- Status
- Pending
- Source link
- https://www.efta.int/eea-lex/32024r1689
What this law does
Providers of general-purpose AI models must draw up and keep up to date the model's technical documentation for the AI Office and national competent authorities, make information and documentation available to providers who intend to integrate the model into their own AI systems, put in place a policy to comply with Union copyright law and in particular to identify and comply with a reservation of rights expressed under Article 4(3) of Directive (EU) 2019/790, and draw up and make publicly available a sufficiently detailed summary of the content used for training, following a template provided by the AI Office.
Paragraph 2 exempts from the first two duties only a model released under a free and open-source licence that allows access, usage, modification and distribution and whose parameters, including the weights, and information on the model architecture and on model usage are made publicly available, and not at all where the model carries systemic risk, so the copyright policy and the training-content summary bind every provider of a general-purpose model.
Article 113(b) applies Chapter V, which contains this Article, from , a year ahead of the Regulation's general application date.
Guidance on this law
How the bodies that enforce this law read it. Guidance binds nobody by itself, so LexLint never raises a finding from it; the duty is this law's.
- Commission Guidelines on the scope of the obligations for providers of general-purpose AI models under the AI Act, the European Commission
When LexLint raises it
When your app profile says your app provides a general-purpose AI model.
Back to the example · Lint your app
Page updated .