Art. 53 and Annexes XI and XII: Obligations for providers of general-purpose AI models under the EU AI Act

Art. 53 places four duties on the provider of a general-purpose AI model. The provider keeps technical documentation with the Annex XI content and gives downstream AI system providers information with the Annex XII content. The provider puts in place a copyright policy that identifies rights reservations. The provider publishes a training-content summary in the AI Office template. In this mapping, eight elements are testable, eight are documentation and eight are process. Regulation (EU) 2026/1744 did not move the GPAI application date 2025-08-02In force or change Art. 53.

Where this applies

2025-08-02In forceGPAI obligations, Art. 53-55
2026-08-02In forceCommission enforcement powers for GPAI (fines, requests for information)

What the text requires

Art. 53 binds the provider of the model, and Art. 54 adds the duty of a provider established outside the Union to appoint an authorised representative. A model with systemic risk carries the additional duties in Art. 55 and the Section 2 elements of Annex XI. Where the model is placed on the market inside an AI system that generates synthetic content, the marking duty in Art. 50(2) falls on the provider of that system.

What the text requires and how each element is verified
Requires	Testable	Documentation	Process
Art. 53(1)(a)The provider draws up the technical documentation of the model. The provider keeps the technical documentation up to date. The documentation includes the training and testing process of the model. The documentation includes the results of the evaluation of the model. The documentation contains, at a minimum, the information set out in Annex XI. The provider gives the documentation to the AI Office and to the national competent authorities on request. Art. 53(1)(b)The provider draws up, keeps up to date and makes available information and documentation to providers of AI systems who intend to integrate the model into their AI systems. The information and documentation enable those providers to have a good understanding of the capabilities and limitations of the model. The information and documentation enable those providers to comply with their obligations under this Regulation. The information and documentation contain, at a minimum, the elements set out in Annex XII. This duty is without prejudice to the protection of intellectual property rights, confidential business information and trade secrets under Union and national law. Art. 53(1)(c)The provider puts in place a policy to comply with Union law on copyright and related rights. The policy identifies and complies with a reservation of rights expressed under Article 4(3) of Directive (EU) 2019/790. The policy uses state-of-the-art technologies for that identification and compliance. Art. 53(1)(d)The provider draws up a sufficiently detailed summary about the content used for training of the model. The provider makes the summary publicly available. The summary follows a template provided by the AI Office. Art. 53(2)The duties in Art. 53(1)(a) and (b) do not apply to providers of AI models released under a free and open-source licence. The licence allows the access, usage, modification and distribution of the model. The parameters, including the weights, the information on the model architecture and the information on model usage are made publicly available. This exception does not apply to general-purpose AI models with systemic risk. Art. 53(3)The provider cooperates as necessary with the Commission and the national competent authorities in the exercise of their competences and powers under this Regulation. Art. 53(4)The provider may rely on a code of practice within the meaning of Art. 56 to demonstrate that the duties in Art. 53(1) are met, until a harmonised standard is published. Compliance with a European harmonised standard grants the provider a presumption of conformity to the extent that the standard covers the duty. No harmonised standard for general-purpose AI models is cited in the Official Journal as of the anchor date, so no presumption of conformity is available under this paragraph today. A provider who does not adhere to an approved code of practice and does not comply with a European harmonised standard demonstrates alternative adequate means for assessment by the Commission. Art. 53(5)The Commission may adopt delegated acts under Art. 97 to detail measurement and calculation methodologies for the documentation in Annex XI, in particular points 2(d) and 2(e) of Annex XI. The purpose is comparable and verifiable documentation. Art. 53(6)The Commission may adopt delegated acts under Art. 97 to amend Annex XI and Annex XII in the light of evolving technological developments. Art. 53(7)Information and documentation obtained under Art. 53, including trade secrets, are treated in accordance with the confidentiality obligations in Art. 78. Annex XI Section 1 point 1The technical documentation contains a general description of the model. The description states the tasks the model is intended to perform and the type and nature of AI systems into which it can be integrated. The description states the acceptable use policies. The description states the date of release and the methods of distribution. The description states the architecture and the number of parameters. The description states the modality and the format of inputs and outputs. The description states the licence. Annex XI Section 1 point 2The technical documentation contains a detailed description of the elements of the model and relevant information on the development process. The description states the technical means required to integrate the model into AI systems. The description states the design specifications of the model and the training process, including training methodologies and techniques, the key design choices with their rationale and assumptions, what the model is designed to optimise for and the relevance of the parameters. The description states information on the data used for training, testing and validation, including the type and provenance of data, the curation methodologies, the number of data points, their scope and main characteristics, how the data was obtained and selected, and the measures to detect unsuitable data sources and biases. The description states the computational resources used to train the model, including the number of floating-point operations, the training time and other relevant details. The description states the known or estimated energy consumption of the model, or an estimate based on the computational resources used where the consumption is unknown. Annex XI Section 2The provider of a general-purpose AI model with systemic risk adds three elements to the technical documentation. The provider adds a detailed description of the evaluation strategies, including the evaluation results, on the basis of available public evaluation protocols and tools or other evaluation methodologies, with the evaluation criteria, the metrics and the methodology on the identification of limitations. The provider adds, where applicable, a detailed description of the measures put in place for internal or external adversarial testing, such as red teaming, and of model adaptations, including alignment and fine-tuning. The provider adds, where applicable, a detailed description of the system architecture that explains how software components build on or feed into each other and integrate into the overall processing. Annex XII point 1The downstream information contains a general description of the model. The description states the tasks the model is intended to perform and the type and nature of AI systems into which it can be integrated. The description states the acceptable use policies. The description states the date of release and the methods of distribution. The description states how the model interacts, or can be used to interact, with hardware or software that is not part of the model, where applicable. The description states the versions of relevant software related to the use of the model, where applicable. The description states the architecture and the number of parameters. The description states the modality and the format of inputs and outputs. The description states the licence. Annex XII point 2The downstream information contains a description of the elements of the model and of the process for its development. The description states the technical means, such as instructions for use, infrastructure and tools, required to integrate the model into AI systems. The description states the modality and format of the inputs and outputs and their maximum size, such as the context window length. The description states information on the data used for training, testing and validation, where applicable, including the type and provenance of the data and the curation methodologies.	Provider status after modification: whether a fine-tuning, continued training or other modification makes the modifier the provider of a new general-purpose AI model, determined against the criteria in the Commission guidelines on GPAI scope, including the compute used for the modification relative to the training compute of the original model Art. 53(2) exception status: whether the licence allows access, usage, modification and distribution, and whether the parameters including the weights, the architecture information and the usage information are publicly accessible at the stated locations Annex XI completeness: every element of Section 1 point 1(a) to (f) and point 2(a) to (e), and of Section 2 for a model with systemic risk, is present in the technical documentation and traceable to a record Annex XII completeness: every element of point 1(a) to (h) and point 2(a) to (c) is present in the information made available to downstream AI system providers Cross-document consistency: the architecture, number of parameters, modality, format, release date, distribution method and licence match between the Annex XI documentation, the Annex XII information and the public model card Declared input and output limits: the modality, format and maximum input and output size stated under Annex XII point 2(b) match the behaviour observed on a sample of inputs at and beyond the stated limits Evaluation result traceability: each evaluation result cited under Art. 53(1)(a) carries the evaluation protocol name, the dataset version and the model version, and each identifier resolves to a stored record Training-content summary structure: the published summary follows the sections of the AI Office template and each section carries content rather than a placeholder	Technical documentation of the model that includes the training and testing process and the results of the evaluation (Art. 53(1)(a)) Information and documentation for downstream AI system providers that describes the capabilities and limitations of the model (Art. 53(1)(b)(i)) Copyright policy that names the state-of-the-art technologies used to identify and comply with rights reservations under Article 4(3) of Directive (EU) 2019/790 (Art. 53(1)(c)) Public summary of the training content in the AI Office template (Art. 53(1)(d)) Record of the number of floating-point operations, the training time and the known or estimated energy consumption with the basis of the estimate (Annex XI Section 1 points 2(d) and 2(e)) For a model with systemic risk, the description of evaluation strategies and results, adversarial testing and model adaptation measures, and system architecture (Annex XI Section 2) Statement of adherence to an approved code of practice, or a record of the alternative adequate means presented for assessment by the Commission (Art. 53(4)) Identification of trade secrets and confidential business information within the documentation handed over, for treatment under Art. 78 (Art. 53(7))	Keep the technical documentation up to date across model versions (Art. 53(1)(a)) Keep the downstream information up to date and make it available to AI system providers who intend to integrate the model (Art. 53(1)(b)) Apply the copyright policy in data collection and training, including the identification of rights reservations, and revise the policy as the state of the art changes (Art. 53(1)(c)) Keep the training-content summary publicly available and revise it when the training content changes (Art. 53(1)(d)) Repeat the provider-status determination and the Art. 53(2) exception check whenever a modification, continued training or a change of licence alters the model (Art. 53(1), Art. 53(2)) Respond to requests for the technical documentation from the AI Office and national competent authorities (Art. 53(1)(a)) Cooperate with the Commission and national competent authorities in the exercise of their powers (Art. 53(3)) Track delegated acts that detail the Annex XI measurement methodologies or amend Annex XI and XII, and update the documentation structure when they apply (Art. 53(5), Art. 53(6))

Metrics and where thresholds come from

Threshold sources, in order of precedence

A harmonised standard cited in the Official Journal (none as of the anchor date)
A JTC 21 draft standard (prEN), with its stage
An ISO/IEC standard or technical report, with its stage
A Commission guideline or code of practice, with its date from the standards data
NIST AI RMF or a sector framework
The state of the art, which the provider justifies and documents

Art. 53 and Annexes XI and XII set no numeric threshold. The annexes list content elements, not values. The only numbers near this article are indicative criteria in the Commission guidelines on GPAI scope: a training-compute criterion for what counts as a general-purpose AI model, stated with a generality condition, and a compute-fraction criterion for when a modification makes the modifier a provider, stated as indicative and open to rebuttal. No harmonised standard for general-purpose AI models is cited in the Official Journal as of the anchor date, so the hierarchy level in use is a Commission guideline and a Code of Practice, each shown with its stage and date from data. The provider justifies its completeness and consistency targets against the model's release context and intended integrations; the determination record step of PR-CLS and the gap record step of PR-11-DOC hold the justification.

Metric sources and their stage
Source and stage	Title	What it provides here	Verified on
`GPAI Code of Practice`Published	General-Purpose AI Code of Practice (transparency, copyright, safety and security chapters)	Transparency chapter: the Model Documentation Form as a field-level structure for the Annex XI and Annex XII content, Measure 1.2 on retention of the documentation for at least ten years for signatories, and Measure 1.3 on what is made available to downstream providers; Copyright chapter Measures 1.1 to 1.5 as the checklist for the Art. 53(1)(c) policy, including lawful access, machine-readable rights reservations such as robots.txt, mitigation of infringing outputs and a point of contact	Verified on 18 September 2026European Commission, The General-Purpose AI Code of Practice
`Commission guidelines on GPAI scope`Published	Guidelines on the scope of obligations for providers of general-purpose AI models under the AI Act	the criteria for what counts as a general-purpose AI model, including the indicative training-compute criterion and its generality condition; the indicative compute-fraction criterion for when a modifier becomes the provider of a new model; and the conditions of the Art. 53(2) free and open-source exception	Verified on 18 September 2026European Commission, Guidelines on the scope of obligations for providers of general-purpose AI models

Procedures

Each protocol is versioned; its page states the derivation from the standards above. PR-CLS v1.0 covers the provider determination after modification and the Art. 53(2) exception check; PR-11-DOC v1.0 covers Annex XI and Annex XII completeness, cross-document consistency and evaluation result traceability.

PR-11-DOC v1.0Annex IV documentation assembly and consistency protocolv1.0
PR-CLS v1.0Classification and role determination protocolv1.0

What an assessor asks for

Annex XI technical documentation (Annex XI Section 1 and, for systemic risk, Section 2)
Annex XII downstream information pack (Art. 53(1)(b))
Copyright policy with rights-reservation handling description (Art. 53(1)(c))
Public training-content summary in the AI Office template (Art. 53(1)(d))
Provider determination record for a modified or fine-tuned model (PR-CLS output)
Open-source exception assessment record (PR-CLS output)
Annex XI and XII completeness and consistency report (PR-11-DOC output)
Code of practice adherence statement or alternative-means record (Art. 53(4))

Common findings

Technical documentation exists as research notes and a model card without a mapping to the Annex XI points
The number of parameters, the architecture and the licence are stated differently in the technical documentation, the downstream information and the public model card
The number of floating-point operations and the training time are estimated after training without the hardware and duration records behind the estimate
The copyright policy states a commitment to respect rights reservations without naming the technologies that read them
The training-content summary is published as free text rather than in the AI Office template
Fine-tuned models are released without a recorded determination of whether the modifier has become the provider

Questions

What does AI Act Article 53 require from a provider?

Art. 53(1) sets four duties for the provider of a general-purpose AI model. The provider keeps technical documentation with at least the Annex XI content available to the AI Office and national competent authorities. The provider gives downstream AI system providers information with at least the Annex XII content. The provider puts in place a copyright policy that identifies rights reservations under Article 4(3) of Directive (EU) 2019/790. The provider publishes a training-content summary in the AI Office template. Art. 53(3) adds a duty to cooperate with the Commission and national competent authorities.

Which GPAI obligations apply from the GPAI date?

The Art. 53 duties apply from the GPAI application date shown on the regulation page. The Digital Omnibus did not change the content of Art. 53 or its application date. Commission enforcement powers under Art. 101 apply from the later date shown on the regulation page. Models with systemic risk carry the additional duties in Art. 55.

What is a GPAI model under the AI Act?

Art. 3(63) defines a general-purpose AI model by significant generality and the capability to perform a wide range of distinct tasks. The Commission guidelines on GPAI scope give an indicative training-compute criterion combined with a generality condition, such as the ability to generate language, text-to-image or text-to-video. The guidelines state that a model can fall outside the criterion and still be a general-purpose AI model where it shows a wide range of capabilities. PR-CLS records the determination with its rationale.

When do GPAI model providers include a company that fine-tunes a model?

The Commission guidelines on GPAI scope treat a modifier as the provider of a new general-purpose AI model when the compute used for the modification exceeds one third of the training compute of the original model. The guidelines present that fraction as an indicative criterion. Below it, the modifier remains a downstream provider and documents the modification rather than the whole model. PR-CLS records the compute figures and the resulting role.

How does the GPAI Code of Practice relate to Article 53?

Art. 53(4) lets a provider rely on a code of practice under Art. 56 to demonstrate that the Art. 53(1) duties are met until a harmonised standard is published. The GPAI Code of Practice has a transparency chapter, a copyright chapter and a safety and security chapter. The transparency chapter supplies the Model Documentation Form that covers the Annex XI and Annex XII content. The copyright chapter supplies measures for the Art. 53(1)(c) policy. No harmonised standard for general-purpose AI models is cited in the Official Journal as of the anchor date.

What changes for GPAI Code of Practice signatories compared with non-signatories?

A signatory demonstrates the Art. 53(1) duties through the measures in the Code. A non-signatory that does not comply with a harmonised standard demonstrates alternative adequate means for assessment by the Commission under Art. 53(4). The underlying duties in Art. 53(1) are the same for both. Signatory status changes the route of evidence.

How we test this

We test the technical documentation and the downstream information against the Annex XI and Annex XII content points and verify consistency between both documents and the public model card. We record the provider determination for a modified or fine-tuned model and prepare the evidence an assessor asks for; the scope limits are stated on what we cannot do.

GPAI evaluation supportPR-15-ROB, PR-CLS, PR-11-DOC

What we cannot do

We do not perform conformity assessments where a notified body is required.
We do not issue CE marking or the EU Declaration of Conformity.
We are not a notified body.
We do not give legal opinions on classification or liability.
We test against requirements, prepare evidence, and structure documentation; compliance decisions remain with the provider or deployer, their notified body, and their counsel.

Read the full statement