Skip to main content
Praxikon
All obligations
Applicablev1.0.0

Article 56: codes of practice for general-purpose AI models

The AI Office encourages and facilitates the drawing up of codes of practice at Union level, and the AI Office and the Board see to it that those codes cover at least the obligations in Articles 53 and 55. Providers of general-purpose AI models and national competent authorities may be invited to participate in the drawing up; civil society organisations, industry, academia and other stakeholders may support the process. Until 27 July 2026 the Commission could approve a code of practice by implementing act and give it general validity within the Union; Article 1, point (21), of Regulation (EU) 2026/1744 replaced paragraph 6 and removed that power. Since then the Commission assesses whether the codes cover the obligations of Articles 53 and 55 and publishes that assessment. The codes were to be ready by 2 May 2025 at the latest; if no code existed by 2 August 2025, or if the AI Office deems one inadequate, the Commission may lay down common rules by implementing acts for the implementation of Articles 53 and 55. For you this is therefore not a separate duty but a route: a code is a voluntary instrument with which you can demonstrate compliance.

Paragraph 1 provides that the AI Office shall encourage and facilitate the drawing up of codes of practice at Union level in order to contribute to the proper application of this Regulation, taking into account international approaches.

Praxikon tracks Article 56: codes of practice for general-purpose AI models under the EU AI Act, checked against the official source on 6 September 2026, citing the source for every statement.

Status
Applicable
Application date
Not recorded
Version
1.0.0
Last reviewed
6 September 2026

Review status: placed against the official source (6 September 2026). Next check due by 5 March 2027. The check date is the knowledge date of this version; no later recheck has been recorded.

From source to evidence

Why this obligation applies, what it asks of you, and what you show for it.

Applies

Applicable · Not recorded

For whom

  • Provider of a GPAI model
  • Provider of an AI system

What you do

Take and record the decision whether you adhere to a code of practice

What you record

Record of the decision on a code of practice

Official source

Article 56(1)-(9)

Who this is relevant to

When this applies

  • Provider of a GPAI model

    A party that places a general-purpose AI model on the Union market.

  • Provider of an AI system

    A party that develops or has an AI system developed and places it on the market under its own name.

  1. 1Arises for the provider of a general-purpose AI model: it may be invited under paragraph 3 to participate in the drawing up of a code of practice, and under paragraph 7 to adhere to a code of practice.
  2. 2Arises for other stakeholders: paragraph 3 names civil society organisations, industry, academia and other relevant stakeholders, such as downstream providers and independent experts, that may support the process.

What the official source establishes

Paragraph 4 provides that the AI Office and the Board shall aim to ensure that the codes of practice clearly set out their specific objectives and contain commitments or measures, including key performance indicators as appropriate, to ensure the achievement of those objectives, and that they take due account of the needs and interests of all interested parties, including affected persons, at Union level. Paragraph 5 provides that the AI Office shall aim to ensure that participants to the codes of practice report regularly to the AI Office on the implementation of the commitments and the measures taken and their outcomes, including as measured against the key performance indicators as appropriate. Key performance indicators and reporting commitments shall reflect differences in size and capacity between various participants. Paragraph 6, as replaced by Article 1, point (21), of Regulation (EU) 2026/1744, provides that the Commission and the Board shall regularly monitor and evaluate the achievement of the objectives of the codes of practice by the participants and their contribution to the proper application of this Regulation. The Commission, taking utmost account of the opinion of the Board, shall assess whether the codes of practice cover the obligations provided for in Articles 53 and 55, and shall regularly monitor and evaluate the achievement of their objectives. The Commission shall publish its assessment of the adequacy of the codes of practice. The power to approve a code of practice by implementing act and give it general validity within the Union, which sat in the second subparagraph of paragraph 6 until 27 July 2026, lapsed with that replacement. Paragraph 7 provides that the AI Office may invite all providers of general-purpose AI models to adhere to the codes of practice. For providers of general-purpose AI models not presenting systemic risks this adherence may be limited to the obligations provided for in Article 53, unless they declare explicitly their interest to join the full code.

Paragraph 8 provides that the AI Office shall, as appropriate, also encourage and facilitate the review and adaptation of the codes of practice, in particular in light of emerging standards. The AI Office shall assist in the assessment of available standards. Paragraph 9 provides that codes of practice shall be ready at the latest by 2 May 2025. The AI Office shall take the necessary steps, including inviting providers pursuant to paragraph 7. The second subparagraph of paragraph 9 provides that if, by 2 August 2025, a code of practice cannot be finalised, or if the AI Office deems it is not adequate following its assessment under paragraph 6 of this Article, the Commission may provide, by means of implementing acts, common rules for the implementation of the obligations provided for in Articles 53 and 55, including the issues set out in paragraph 2 of this Article. Those implementing acts shall be adopted in accordance with the examination procedure referred to in Article 98(2).

Our interpretation

The official source remains authoritative. This general interpretation is not legal advice.

A code of practice is a voluntary instrument, and that word does all the work here. Signing creates no new obligation: your obligations remain those of Articles 53 and 55, and Article 56 changes nothing about them. Not signing is not an infringement either, because paragraphs 3 and 7 speak of inviting and not of requiring. What does shift is the burden of proof. A provider adhering to a code can point to a shared elaboration, assessed by the AI Office and the Board, when a regulator asks how it keeps its documentation up to date, how detailed its summary about the training content is, or how it assesses and manages systemic risk. A provider that does not sign has to write that elaboration itself and defend it itself, up to and including the question why its own approach is at least as good. That is not a legal difference in the norm, but it is a large difference in what is on the table when something is asked. Two things that are often conflated here. First: paragraph 7 allows a provider without a systemic-risk model to limit itself to the obligations in Article 53, unless it explicitly declares its interest in the full code. Partial participation is therefore an expressly foreseen choice and not half-heartedness. Second: paragraph 9 puts a stick behind the door that does not rest on you but does reach you. If no code comes about, or if the AI Office deems it inadequate, the Commission may lay down common rules by implementing acts for the implementation of Articles 53 and 55. Those rules, unlike a code, are not voluntary.

Precision is in order about the state of play, because this is where the market overstates the most. It is established that the General-Purpose AI Code of Practice was published on 10 July 2025; that code is archived in this repository as three chapter PDF files, on transparency, copyright and safety and security, and it is present as a source record in this knowledge base. Whether an approving implementing act followed was not established for the period up to 27 July 2026, and after that date it is no longer the right question: paragraph 6 no longer carries that power. Since then the question is whether the Commission has published its assessment of adequacy. That too is not established here, and for as long as that is the case you must not read anywhere that the code has been approved or has general validity within the Union. That distinction is not a formality: a published code is a text you can voluntarily adhere to, a code approved by implementing act is on top of that an instrument with general validity in the Union. Anyone mistaking the first for the second overestimates what a signature buys and underestimates what they still have to record themselves. The content of the three chapters was also not read in this build, so nothing in this object says anything about what exactly is in them. In practice that means the following. Check the approval status yourself and by date before you rely on the code in a conversation with a regulator or a customer, and record which version of the code and which chapter your adherence relates to. A code may be reviewed and adapted under paragraph 8, in particular in light of emerging standards, so a reliance on the code without a version reference ages on its own.

What you can do now

Treat the question whether you adhere to a code of practice as a decision that is taken and recorded, not as something that happens by itself. Record per model: do you adhere to a code, to which version and to which chapter, and if not, which elaboration of your own you apply instead for the issues named in paragraph 2, namely keeping the information in Article 53(1), points (a) and (b), up to date, the level of detail of the summary about the training content, and, where you offer a systemic-risk model, the identification and management of systemic risk. Decide explicitly whether under paragraph 7 the obligations in Article 53 are enough for you or whether you join the full code, and note that choice with a date and an authorised signatory. Before you rely on a code externally, check whether the Commission has published its assessment of adequacy under the amended paragraph 6, and claim no more than you can point at that moment. Do not ask for an approving implementing act: that power has not existed since 27 July 2026, and asking for it is asking for a decision nobody can take any more. Finally, put a review moment in your own calendar: paragraph 8 provides for review and adaptation of codes in light of emerging standards, and paragraph 9 allows the Commission to lay down common rules where a code fails to appear or is deemed inadequate, and both change what you are relying on without anyone calling you.

  1. 01

    Take and record the decision whether you adhere to a code of practice

    Determine per general-purpose AI model whether you adhere to a code of practice, to which version and which chapter, whether under paragraph 7 the obligations in Article 53 suffice for you, and which elaboration of your own you apply for the issues in paragraph 2 where you do not join.

What to retain

Record of the decision on a code of practice

Per model: the decision whether or not to adhere to a code of practice, the version and chapter it relates to, the date and the authorised signatory, whether adherence was limited under paragraph 7 to the obligations in Article 53, and, where the decision is negative, the elaboration of your own for the issues in paragraph 2.

Control and reassessment

  • Review moment on your reliance on a code of practice

    The control that periodically rechecks a reliance on a code of practice: does the version you rely on still stand, has the code been reviewed or adapted under paragraph 8, has the Commission published its assessment of adequacy under paragraph 6 as replaced by Regulation (EU) 2026/1744, and has the Commission laid down common rules under paragraph 9.

Public tools

Conditions and exceptions

  • Paragraph 7 provides that for providers of general-purpose AI models not presenting systemic risks this adherence may be limited to the obligations provided for in Article 53, unless they declare explicitly their interest to join the full code. A provider without a systemic-risk model therefore does not have to sign up to the systemic-risk part in order to rely on the code.
  • Article 56 does not impose a separate obligation on the provider. Paragraphs 3 and 7 speak of inviting, not of requiring, and the obligations themselves remain those of Articles 53 and 55. Not signing is therefore not an infringement of Article 56.

Official sources and locators

  • EU Artificial Intelligence Act 2024/1689

    European Parliament and Council | original-oj-2024-07-12

    Source locator: Article 56(1)-(9)

  • Digital Omnibus on AI 2026/1744

    European Parliament and Council | official-journal-2026-07-24

    Source locator: Article 1, point (21), replacing Article 56(6)

  • General-Purpose AI Code of Practice

    European Commission / AI Office | published-2025-07-10

    Source locator: Code of Practice for General-Purpose AI, 10 July 2025

Referring to this object

Citation block

Copy this reference into your advice, article or file. The identifier, the version and the hash keep the statement findable later, even once the dataset has moved on.

Reference

Praxikon, "Article 56: codes of practice for general-purpose AI models",
praxikon:eu:ai-act:obligation:article-56-gpai-codes-of-practice@1.0.0,
dataset praxikon:sys:registry:dataset:ai-act-implementation-graph 2.2.0 (schema 1.5.0),
effective_at 2025-08-02T00:00:00.000Z, known_at 2026-09-06T00:00:00.000Z,
sha256 f6e6afc9742736b52408f8c38c9435fc8be751641392f5d0b5d57a34afee4c1c,
https://www.praxikon.com/en/verplichtingen/article-56-gpai-codes-of-practice
(https://www.praxikon.com/api/v1/obligations?id=praxikon%3Aeu%3Aai-act%3Aobligation%3Aarticle-56-gpai-codes-of-practice&effective_at=2025-08-02&known_at=2026-09-06&lang=en, accessed 2026-09-15)

Short form

praxikon:eu:ai-act:obligation:article-56-gpai-codes-of-practice@1.0.0 (sha256 f6e6afc9)

BibTeX

@misc{praxikon-eu-ai-act-obligation-article-56-gpai-codes-of-practice-1-0-0,
  author       = {{Praxikon}},
  title        = {Article 56: codes of practice for general-purpose AI models},
  year         = {2026},
  version      = {1.0.0},
  number       = {praxikon:eu:ai-act:obligation:article-56-gpai-codes-of-practice},
  howpublished = {AI Act Change \& Evidence Graph, dataset 2.2.0, schema 1.5.0},
  note         = {effective_at 2025-08-02T00:00:00.000Z; known_at 2026-09-06T00:00:00.000Z; sha256 f6e6afc9742736b52408f8c38c9435fc8be751641392f5d0b5d57a34afee4c1c},
  url          = {https://www.praxikon.com/en/verplichtingen/article-56-gpai-codes-of-practice},
  urldate      = {2026-09-15},
  language     = {en}
}

CSL JSON

[
  {
    "id": "praxikon:eu:ai-act:obligation:article-56-gpai-codes-of-practice@1.0.0",
    "type": "dataset",
    "title": "Article 56: codes of practice for general-purpose AI models",
    "container-title": "AI Act Change & Evidence Graph",
    "publisher": "Praxikon",
    "version": "1.0.0",
    "number": "praxikon:eu:ai-act:obligation:article-56-gpai-codes-of-practice",
    "URL": "https://www.praxikon.com/en/verplichtingen/article-56-gpai-codes-of-practice",
    "language": "en",
    "issued": {
      "date-parts": [
        [
          2026,
          9,
          6
        ]
      ]
    },
    "accessed": {
      "date-parts": [
        [
          2026,
          9,
          15
        ]
      ]
    },
    "note": "dataset praxikon:sys:registry:dataset:ai-act-implementation-graph 2.2.0; schema 1.5.0; effective_at 2025-08-02T00:00:00.000Z; known_at 2026-09-06T00:00:00.000Z; sha256 f6e6afc9742736b52408f8c38c9435fc8be751641392f5d0b5d57a34afee4c1c; retrieved_from https://www.praxikon.com/api/v1/obligations?id=praxikon%3Aeu%3Aai-act%3Aobligation%3Aarticle-56-gpai-codes-of-practice&effective_at=2025-08-02&known_at=2026-09-06&lang=en; licence https://www.praxikon.com/nl/legal/terms"
  }
]

How to verify a reference later is set out in the methodology. Terms

Version history

  1. v1.0.0

    2 August 2025

    Article 56: codes of practice for general-purpose AI models

    The AI Office encourages and facilitates the drawing up of codes of practice at Union level, and the AI Office and the Board see to it that those codes cover at least the obligations in Articles 53 and 55. Providers of general-purpose AI models and national competent authorities may be invited to participate in the drawing up; civil society organisations, industry, academia and other stakeholders may support the process. Until 27 July 2026 the Commission could approve a code of practice by implementing act and give it general validity within the Union; Article 1, point (21), of Regulation (EU) 2026/1744 replaced paragraph 6 and removed that power. Since then the Commission assesses whether the codes cover the obligations of Articles 53 and 55 and publishes that assessment. The codes were to be ready by 2 May 2025 at the latest; if no code existed by 2 August 2025, or if the AI Office deems one inadequate, the Commission may lay down common rules by implementing acts for the implementation of Articles 53 and 55. For you this is therefore not a separate duty but a route: a code is a voluntary instrument with which you can demonstrate compliance.

Corrections to this obligation

No substantive correction to this object has been recorded.

Open the correction log
Zahed Ashkara, jurist and freelance AI & Privacy Consultant

Behind this page

Zahed Ashkara

Freelance AI & Privacy Consultant, jurist

Help with implementation

Zahed Ashkara, jurist and freelance AI & Privacy Consultant, supports implementation with your team through Embed AI.

View AI governance at Embed AI

For AI agents and integrations

This page and the machine output derive from the same versioned object. Use the API for deterministic filters by role, topic and time.