Topic
GPAI and generative models
General-purpose AI models carry their own obligations on documentation, copyright and systemic risk. The Commission's enforcement powers have been active since 2 August 2026. The hardest question in practice is role determination: adapting or passing on a model can make you a provider yourself.
Official guidance on this topic
11- Guidelines on the scope of obligations for providers of general-purpose AI models under the AI ActNon-binding guidelines explaining when a model constitutes a general-purpose AI model, with an indicative threshold of more than 10^23 floating point operations training compute in combination with the capacity to generate language, image or video. They determine who qualifies as a provider, when a party that fine-tunes becomes a provider itself, when a provider outside the Union must appoint an authorised representative, and under what conditions open source is exempted. Available in all 24 EU languages and linked to the application of the general-purpose AI rules from 2 August 2025.PublishedGuidelinesEuropean Commission18 Jul 2025
- The General-Purpose AI Code of PracticeThe General-Purpose AI code of practice was delivered on 10 July 2025 by thirteen independent experts following input from more than one thousand stakeholders. The code consists of three chapters: Transparency and Copyright for all providers of general-purpose AI models, and Safety and Security only for providers of models with systemic risk. Signature is voluntary, but the Commission and the AI Board confirmed the code on 1 August 2025 as an adequate instrument, allowing signatories to demonstrate their obligations with less administrative burden. Those who do not sign must substantiate compliance in another, equally adequate manner.PublishedCode of practiceEuropean Commission / AI Office10 Jul 2025
- Report for Serious Incidents under the AI Act (General-Purpose AI Models with Systemic Risk)Standard form (DOCX and PDF) by which providers of general-purpose AI models with systemic risk report serious incidents to the AI Office and, where necessary, to national competent authorities. It operationalises commitment 9 from the safety chapter of the general-purpose AI Code of Practice and shows what level of detail the AI Office expects on cause, impact and measures taken. Not to be confused with the draft document for Article 73 (high-risk systems): different legal basis, different status.PublishedTemplateEuropean Commission / AI Office4 Nov 2025
- Explanatory Notice and Template for the Public Summary of Training Content for general-purpose AI modelsThe model prescribed by the Commission by which providers of general-purpose AI models publicly summarise what data they used for training. Use of this format is mandatory under Article 53, paragraph 1, point d. The summary has three parts: general information about provider and model, a list of data sources (public and private datasets, scraped data, user data, synthetic data) including top-level domain names, and processing aspects including the removal of illegal content. Updates occur every six months or earlier on material changes. Note: the downloadable version was revised after 24 July 2025, so always cite the version date.PublishedTemplateEuropean Commission / AI Office24 Jul 2025
- Questions and answers on the Code of Practice for General-Purpose AIQ&A on the code of practice for general-purpose AI models. The Commission emphasises that the code is voluntary and creates no new obligations, but serves as a tool to comply with the existing obligations in the regulation. The code has three chapters: transparency and copyright for all general-purpose AI providers, and safety and security only for providers of the most advanced models with systemic risk. It also clarifies how the transparency rules for general-purpose AI models relate to the transparency obligations of Article 50 for AI systems. The date 20 July 2026 is the last updated date on the page.PublishedQ&AEuropean Commission, DG CONNECT20 Jul 2026
- Guidelines on obligations for General-Purpose AI providers - Questions & AnswersQ&A on the guidelines for providers of general-purpose AI models. The document makes concrete the indicative computational threshold of 10^23 FLOP as a starting point for the question of whether something is a general-purpose AI model, with the nuance that a model above that threshold might nevertheless fall outside the definition if it lacks genuine generality. It further explains when a party that adapts an existing model itself becomes a provider, with the rule of thumb that more than one third of the original training computational capacity is used. The scope of the open-source exception is also defined. The date 11 November 2025 appears as last updated at the bottom of the page.PublishedQ&AEuropean Commission, DG CONNECT11 Nov 2025
- General-Purpose AI Models in the AI Act - Questions & AnswersOfficial Q&A on general-purpose AI models. The page explains what a general-purpose AI model is, when a model is classified as a model with systemic risk, what obligations providers have regarding technical documentation, information to users, copyright policy and a summary of training data, and what applies in the case of open-source release or fine-tuning. The role and powers of the AI Office are also covered. The obligations apply from 2 August 2025, with a transitional period until 2 August 2027 for models already on the market before that date. The date 9 September 2025 is the last updated date on the page.PublishedQ&AEuropean Commission, DG CONNECT9 Sept 2025
- Commission Opinion on the assessment of the General-Purpose AI Code of PracticeOn 1 August 2025, the Commission and the AI Board formally confirmed that the general-purpose AI code of practice is an adequate voluntary instrument to demonstrate compliance with the AI Regulation. This decision is the legal link that converts the voluntary code into a usable means of proof towards the AI Office. Without this adequacy confirmation, signature would not carry the same supervisory weight.PublishedOpinionEuropean Commission1 Aug 2025
- Opinion 28/2024 on certain data protection aspects related to the processing of personal data in the context of AI modelsAt the request of the Irish supervisor, the EDPB answers four questions on data protection law and AI models. A model trained on personal data is not automatically anonymous: supervisors test case-by-case whether direct or probabilistic extraction and obtaining personal data via prompts are negligible. The Opinion elaborates the three-step test for legitimate interest for the development and deployment phases, including reasonable expectations in web scraping and mitigating measures, and describes three scenarios in which unlawful processing during development carries through into subsequent use. Adopted 17 December 2024, published 18 December 2024.PublishedOpinionEDPB (European Data Protection Board)17 Dec 2024
- AI Regulation timeline (Algoritmekader)Dutch timeline from BZK showing which requirements apply on each effective date, linked to numbered requirements in the Algorithm Framework. Living document, last updated 3 August 2026, incorporating the postponement: new standalone high-risk systems from 2 December 2027, high-risk in products from 2 August 2028, with an exception until 2030 for systems already in use in government organisations on 2 December 2027. Warning: the timeline is incomplete, as 2 August 2026 is missing as a milestone for the transparency obligations of Article 50. Do not use as a complete timeline source.PublishedNational guidanceMinistry of the Interior and Kingdom Relations (Algorithm Framework)17 Jul 2025
- Guidance on generative AI and the GDPRThe first Dutch interpretation of the GDPR for the development and deployment of generative AI models, intended for organisations that build such models or are responsible for their use. The guidance addresses the legal basis for processing in detail and also works out further GDPR requirements for development and deployment. It builds on the earlier opinion from the Dutch Data Protection Authority on generative AI and refers to the obligations that the AI Act imposes on general-purpose AI model providers. The document is dated June 2026 internally and was published on 13 July 2026.PublishedGuidelinesDutch Data Protection Authority (AP)13 Jul 2026