With the entry into force of the Artificial Intelligence (AI) Act, the complex process of developing harmonised European standards was initiated in order to make the obligations it sets out operational.
In this context, the Codes of Practice (provided for in Article 56 of the Regulation) are a central tool for ensuring compliance with the obligations set out in the AI Act, especially in the interim period between the entry into force of the obligations for general purpose AI model providers (GPAI) (in August 2025) and the adoption of standards (no later than August 2027). For this reason, the codes of practice must be ready by 2 May 2025 at the latest. Although they are not legally binding, compliance with the guidelines set out in these Codes will serve as a ‘presumption of conformity’ with GPAI model provider obligations until more robust standards come into force.
The Codes are being developed through a complex process involving the various stakeholders, individually or through working groups, including Member State representatives, independent experts, academics, AI model providers, members of civil society and others.
The EU AI Service and the Committee assess the suitability of the codes, which can then be approved by the Commission by means of an implementing act. If a code of practice cannot be finalised by 2 August 2025, or if the AI Service considers that it is not appropriate, it will be up to the Commission to establish, by means of implementing acts, common rules for the application of the AI Regulation.
This is the context for the publication, on 19 December 2024, of the second draft of the general-purpose AI Code of Practice, drawn up by independent experts on the basis of feedback received on the first version, published on 14 November 2024.
This version reflects the opinions and reflections of stakeholders who took part in the Code of Practice working groups meetings as well as the seminars attended by suppliers, Member State representatives and European and international observers. Written comments on the first version submitted via a specific platform and suggestions made by suppliers of general-purpose AI models presented at specific workshops have also been taken into account.
The announced aim is to adopt a ‘future-proof’ Code, to be applied to models launched after 2 August 2025, when the new rules on general-purpose AI models come into force. It is intended that the Code will play the role of a guidance document for providers of general-purpose AI models in demonstrating compliance with the AI Act throughout the model lifecycle.
In this version of the Code, the intention has been to clearly indicate the obligations arising from the AI Act that are regulated by it through concrete measures. In addition, the structure of the Code has been adapted to outline objectives, commitments and measures (replacing the structure of the previous draft which consisted of objectives, measures and sub-measures) and to include preliminary examples of key performance indicators (KPIs).
The first part of the draft Code presented specifies transparency and copyright obligations for all providers of general-purpose AI models, with significant exemptions for providers of certain open source models, in accordance with the AI Act.
The second part of the Code has a narrower scope of application, only being relevant to the small number of providers of more advanced general-purpose AI models that may pose systemic risks, according to the classification criteria set out in Article 51 of the AI Act. In this case, the Code defines measures for assessing and mitigating systemic risk, including model assessments, incident reporting and cybersecurity obligations.
The second version of the Code was in public consultation until 15 January 2025, and will also be subject to hearings with experts in the various working groups established and meetings with general-purpose AI model providers and member state representatives. It is planned that the third draft Code of Practice will be made public in the week of 17 February 2025.
#AIAct #GPAI #Codeofpratice #General-PurposeAImodels #seconddraft
*Image source: Commission website
