Explore our collection of PDF documents and enrich your knowledge now!
back
Code of practice for general-purpose AI models: final text remains outstanding!
Article IT and Data Protection | 24/06/25 | 3 min. | Mahasti Razavi Eden Gall Benjamin Fontani
The fourth and final draft of the code of practice for general-purpose artificial intelligence models, which was due for publication on May 2, 2025, remains outstanding.
Pursuant to Article 56 of Regulation (EU) 2024/1689 on Artificial Intelligence (“AI Act”), a non-binding code of practice, developed by a multi-stakeholder expert group supported by the AI Office, was to be published no later than May 2, 2025. The code of practice is intended to guide providers of general-purpose artificial intelligence models in implementing the new obligations imposed by the AI Act and to serve as a reference framework for assessment when such providers choose to demonstrate compliance through this voluntary mechanism.
The process was formally initiated on September 30, 2024, when the AI Office convened nearly 1,000 stakeholders from a wide range of sectors for an inaugural event highlighting the cross-cutting scope and complexity of the challenges at stake.
Subsequently, three draft versions of the code of practice were released on November 14, 2024, December 19, 2024, and March 11, 2025. The third draft presented a streamlined set of general commitments focused on transparency and copyright safeguards, along with detailed operational measures to facilitate their implementation. For models classified as general-purpose artificial intelligence models that could pose systemic risks, the last draft introduced enhanced requirements, including comprehensive risk assessments, mitigation measures, and security reporting obligations. It also proposed a standardized model card accompanying each general-purpose artificial intelligence model, disclosing its capabilities, limitations, and the key characteristics of its training data.
Despite these developments, a final consolidated version of the code of practice had not been issued as of the deadline of May 2, 2025. Although the AI Office has not provided an official explanation, reports from specialized media point to significant disagreements among stakeholders, with some industry actors expressing concern over the technical feasibility and economic impact of the proposed measures.
In an official policy opinion issued on May 14, 2025, the European Affairs Committee of the French Senate described a "widespread dissatisfaction" with the ongoing process. The Committee criticized the third draft of the code of practice for its lack of ambition, particularly with respect to copyright protections, and urged the European Commission to resist pressure from certain providers of general-purpose artificial intelligence models to ensure robust and consistent enforcement of the AI Act.
The consequences of this delay are material for affected providers who are required to comply with the AI Act’s specific obligations regarding general-purpose artificial intelligence as of August 2, 2025. In this regard, Article 56 expressly provides that, if no code of practice is adopted by that date, the European Commission may adopt implementing acts to establish a uniform and binding compliance framework applicable to providers of general-purpose artificial intelligence models.
In this context, it remains uncertain whether the various stakeholders will succeed in reaching an agreement and adopting a broadly supported code of practice before August 2, 2025, or whether the Commission will be compelled to unilaterally impose a framework through an implementing act.
In any event, while the core obligations applicable to providers of general-purpose artificial intelligence models under the AI Act will enter into force in the near future, the penalty provisions set forth in Article 101 will not become enforceable until August 2, 2026.