Skip to content
ContentLora

    Tip: press / anywhere to search.

    concept

    General-Purpose AI Code of Practice (EU)

    Also known as GPAI Code of Practice, EU AI code of practice, General-Purpose AI Code, GPAI code

    The General-Purpose AI Code of Practice is a voluntary tool, written by independent experts, that helps providers of general-purpose AI models comply with the EU AI Act.[1] The Commission published it on 10 July 2025 with chapters on transparency, copyright, and safety and security.[2] As of October 2026, 22 providers had signed the full code, including Anthropic, Google, Microsoft, Mistral AI and OpenAI; xAI signed only the safety chapter, and Meta declined.[3][4][5]

    Editor reviewedStrict sourcingUpdated Wiki22 claims · 91% primaryAI safety and alignment, Tech policy, Artificial intelligence
    Key facts

    This page covers the code for general-purpose AI models. A separate code on marking and labelling AI-generated content, linked to the Act’s transparency rules, is outside its scope. For the law itself, read the EU AI Act explained; for dates, read EU AI Act timeline.

    The General-Purpose AI Code of Practice is a voluntary tool for companies that build large AI models used across many tasks. Independent experts wrote it in a multi-stakeholder process to help providers comply with the AI Act’s obligations for general-purpose AI models.[1] The Commission published it on 10 July 2025.[2]

    How it relates to the AI Act

    The AI Act requires providers of general-purpose AI models to keep technical documentation, adopt an EU copyright policy and publish a summary of training content.[6] Providers of models with systemic risk must also evaluate their models, assess and mitigate systemic risks, report serious incidents and secure their models.[7] These obligations have applied since 2 August 2025.[8]

    Article 56 of the Act tasks the AI Office with encouraging codes of practice. Providers may rely on such a code to show compliance until a harmonised standard is published. Providers that do not adhere to an approved code or standard must show the Commission alternative adequate means of compliance.[9] On 1 August 2025 the Commission and the AI Board confirmed through adequacy decisions that this code is an adequate tool for that purpose.[10] The Commission says signatories get less administrative burden and more legal certainty than if they proved compliance in other ways.[11]

    The code was one of three instruments the Commission published for model providers in July 2025, alongside guidelines on the scope of the obligations and a template for the training-content summary.[12] The template asks providers to describe their data sources, including large datasets and top domain names.[13] Since 2 August 2026 the AI Office can enforce the model rules, including by requesting documentation, evaluating models and issuing fines.[14]

    The three chapters

    The code has three separately authored chapters. The Transparency and Copyright chapters give every general-purpose AI provider a way to show compliance with Article 53. The Safety and Security chapter applies only to the few providers of models with systemic risk under Article 55.[15]

    • Transparency comes with a Model Documentation Form for recording the information the Act requires.[16]
    • Copyright offers practical ways to meet the obligation to have a policy that complies with EU copyright law.[16]
    • Safety and Security sets out state-of-the-art practices for managing systemic risks from the most advanced models.[16]

    How it was drafted

    The AI Office opened a call to take part in drafting on 30 July 2024.[17] Independent experts chaired the process, which involved nearly 1,000 stakeholders, member-state representatives and international observers, working in four groups.[18] The second draft appeared on 19 December 2024 and the third on 11 March 2025, and the Commission received the final text on 10 July 2025.[19] A consultation on the training-data template drew more than 430 responses.[13]

    Who signed

    The Commission’s list, last updated on 7 October 2026, names 22 providers that signed the full code. They include Anthropic, Amazon, Google, IBM, Microsoft, Mistral AI and OpenAI, as well as Aleph Alpha, Black Forest Labs, Cohere, ServiceNow and WRITER.[3] xAI signed only the Safety and Security chapter, so it must show compliance with the transparency and copyright obligations by other means.[4] Signatories have formed a Signatory Taskforce chaired by the AI Office.[20]

    Meta said on 18 July 2025 that it would not sign. Its chief global affairs officer, Joel Kaplan, said the code created legal uncertainty for model developers and went far beyond the scope of the AI Act.[5] Around the same time, industry groups and companies including Mistral, ASML and Airbus urged the EU to delay the Act or the code.[21] The model rules nonetheless took effect on 2 August 2025 as scheduled.[22]

    Developing in AI & Computing

    Questions readers ask

    Is the code of practice legally binding?

    No. It is a voluntary tool. Providers that sign and follow it can use it to show they meet the AI Act's obligations for general-purpose AI models, which are binding.[1][11]

    What happens to companies that do not sign?

    They still have to meet the AI Act's obligations, and must show the Commission alternative adequate means of compliance.[9]

    Which companies signed the code?

    The Commission's list, updated on 7 October 2026, names 22 providers that signed the whole code, including Amazon, Anthropic, Google, IBM, Microsoft, Mistral AI and OpenAI. xAI signed only the Safety and Security chapter.[3][4]

    Why did Meta refuse to sign?

    Meta's Joel Kaplan said in July 2025 that the code introduced legal uncertainties for model developers and measures going far beyond the scope of the AI Act.[5]

    Does every provider need the Safety and Security chapter?

    No. It applies only to the small number of providers of models with systemic risk under Article 55. The Transparency and Copyright chapters are for all general-purpose AI providers.[15]

    Sources

    · 22 claims from 5 publishers

    Each numbered claim is a statement we checked against the sources listed with it. Tap a number in the text to see its sources in place.

    1. [1]

      The General-Purpose AI Code of Practice is a voluntary tool prepared by independent experts in a multi-stakeholder process to help providers comply with the AI Act's obligations for general-purpose AI models. confirmedas of 2026-10-07

    2. [2]

      The European Commission published the General-Purpose AI Code of Practice on 10 July 2025, with chapters on transparency, copyright, and safety and security; the safety and security chapter applies only to providers of general-purpose models with systemic risk. confirmedas of 2025-07-10

    3. [3]

      As of the Commission's list updated on 7 October 2026, 22 providers had signed the full code, including Amazon, Anthropic, Google, IBM, Microsoft, Mistral AI and OpenAI, along with Aleph Alpha, Black Forest Labs, Cohere, ServiceNow and WRITER. confirmedas of 2026-10-07

    4. [4]

      xAI signed only the code's Safety and Security chapter, so it must show compliance with the transparency and copyright obligations by other adequate means. confirmedas of 2026-10-07

    5. [5]

      On 18 July 2025 Meta said it would not sign the code; its chief global affairs officer, Joel Kaplan, said the code introduced legal uncertainties for model developers and measures going far beyond the scope of the AI Act. confirmedas of 2025-07-18

    6. [6]

      Providers of general-purpose AI models must keep technical documentation, give information to downstream providers, put in place a policy to comply with EU copyright law, and publish a sufficiently detailed summary of the content used for training. confirmedas of 2024-07-12

    7. [7]

      Providers of general-purpose AI models with systemic risk must additionally evaluate their models, including adversarial testing, assess and mitigate systemic risks at EU level, track and report serious incidents, and ensure adequate cybersecurity. confirmedas of 2024-07-12

    8. [8]

      The AI Act's rules for general-purpose AI models and its governance rules applied from 2 August 2025. confirmedas of 2025-08-02

    9. [9]

      Article 56 tasks the AI Office with encouraging and facilitating codes of practice; providers of general-purpose AI models may rely on such codes to demonstrate compliance until a harmonised standard is published; those that do not adhere to an approved code or a harmonised standard must show alternative adequate means of compliance. confirmedas of 2024-07-12

    10. [10]

      The Commission and the AI Board confirmed on 1 August 2025, through adequacy decisions, that the code is an adequate voluntary tool for providers to demonstrate compliance with the AI Act. confirmedas of 2025-08-01

    11. [11]

      According to the Commission, signatories that adhere to the code can show they comply with the AI Act with less administrative burden and more legal certainty than through other methods. confirmedas of 2026-10-07

    12. [12]

      In July 2025 the Commission published three instruments for general-purpose AI providers - guidelines on the scope of the obligations, the code of practice, and a template for the public summary of training content. confirmedas of 2025-07-31

    13. [13]

      The Commission adopted a template for the public summary of training data in July 2025, after a consultation that drew more than 430 responses; it asks providers to describe their data sources, including large datasets and top domain names. confirmedas of 2025-07-31

    14. [14]

      From 2 August 2026 the AI Office and member-state authorities are responsible for implementing and enforcing the Act, and the AI Office can request technical documentation, evaluate models, require corrective measures and fine providers of general-purpose AI models. confirmedas of 2026-08-02

    15. [15]

      The code has three separately authored chapters; the Transparency and Copyright chapters give all general-purpose AI providers a way to show compliance with Article 53, while the Safety and Security chapter is relevant only to providers of models with systemic risk under Article 55. confirmedas of 2026-10-07

    16. [16]

      The Transparency chapter includes a Model Documentation Form, the Copyright chapter offers practical ways to meet the obligation to have an EU copyright policy, and the Safety and Security chapter sets out state-of-the-art practices for managing systemic risks. confirmedas of 2026-10-07

    17. [17]

      The AI Office launched the call to participate in drawing up the code on 30 July 2024. confirmedas of 2024-07-30

    18. [18]

      The drafting process, facilitated by the AI Office, was chaired by independent experts and involved nearly 1,000 stakeholders, member-state representatives and international observers, organised in four working groups. confirmedas of 2025-07-10

    19. [19]

      The AI Office published the second draft of the code on 19 December 2024 and the third draft on 11 March 2025, and received the final version on 10 July 2025 after a closing plenary on 3 July. confirmedas of 2025-07-10

    20. [20]

      Signatories have set up a Signatory Taskforce, chaired by the AI Office, to support a coherent application of the code. confirmedas of 2026-10-07

    21. [21]

      In mid-2025 groups representing large US tech companies and European companies including Mistral, ASML and Airbus urged the EU to delay the AI Act or the code. confirmedas of 2025-07-18

    22. [22]

      In July 2025 a Commission spokesperson reiterated that the general-purpose AI rules would take effect on 2 August 2025, and they did. confirmedas of 2025-08-02

    Revision history (1)
    1. Page created.

    Created Nov 7, 2026. Last reviewed by an editor on Nov 8, 2026. Next scheduled review: Feb 7, 2027.

    Cite this page

    "General-Purpose AI Code of Practice (EU)." ContentLora, updated Nov 7, 2026. https://contentlora.com/wiki/general-purpose-ai-code-of-practice

    Spotted an error? Suggest a correction or emailcorrections@contentlora.com.