The Artificial Intelligence Act is in force, and general-purpose model documentation duties start on 2 August 2025

2024-08-01EU Policy

Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), OJ L, 2024/1689, 12 July 2024, in force 1 August 2024 under Article 113

The Artificial Intelligence Act entered into force on 1 August 2024, the twentieth day after publication in the Official Journal. Article 113 sets the application dates: general application from 2 August 2026; the Chapter I general provisions and Chapter II prohibited practices from 2 February 2025; the Chapter V obligations for general-purpose artificial intelligence models, with Chapter III Section 4 on notifying authorities, Chapter VII on governance, Chapter XII on penalties except Article 101, and Article 78 on confidentiality, from 2 August 2025; and Article 6(1), the high-risk classification for products under listed harmonisation legislation, from 2 August 2027. Article 53(1) requires every provider of a general-purpose model to draw up and keep up to date technical documentation covering training, testing and evaluation with at least the Annex XI content, to supply downstream system providers with the Annex XII information while preserving intellectual property rights and trade secrets, to adopt a policy to comply with Union copyright law including the text and data mining opt-outs under Article 4(3) of Directive (EU) 2019/790, and to publish a sufficiently detailed summary of the content used for training on a template from the AI Office. Article 53(2) exempts models released under a free and open-source licence from the first two duties unless the model presents systemic risk, and Article 53(7) routes trade secrets in the documentation to the Article 78 confidentiality regime.

What it changesTwo of the Article 53 duties are patent decisions with a date. The public training-content summary will put on the record the data sources, filtering and augmentation choices that some teams hold as trade secrets, and secrecy ends at publication, so any data-pipeline technique a company wants to protect must be filed before the summary describes it, and the summary should be drafted with patent counsel reading it for what it gives away. The technical documentation on training and testing is a written description of the model's construction; it is confidential to the regulator, but it is the natural source document for European priority filings on training methods, and writing it with claim support in mind costs nothing extra. For a company shipping a model into the European Union the schedule is now fixed: audit the data pipeline for trade-secret-or-patent decisions before 2 August 2025, file what will be patented before the summary is published, and treat the open-weight exemption as applying only to models without systemic risk.