General-Purpose Artificial Intelligence Code of Practice
Whilst many of us were happily taking a well-deserved vacation, on the 10 July the European Commission (EC) was busy publishing its voluntary Code of Practice on General-Purpose AI. This Code is intended to guide providers of general-purpose AI (GPAI) models in complying with the EU’s AI Act, focusing on safety, transparency, and copyright obligations. It addresses the AI Act’s regulatory framework for GPAI models, effective from August 2, 2025, and aims to reduce administrative burdens while increasing legal certainty for AI providers.
GPAI providers are entities (natural or legal persons, public authorities, agencies, or other bodies) that develop a General Purpose AI model or have one developed and then place it on the market or put it into service under their own name or trademark. Although it is not highly likely to happen due to the resources required and regulatory burden imposed, service providers operating in any sector (be it financial services, gaming, transport, medical, legal, employment, or any other), who develop a GPAI model in-house, or outsource the development of such a model, and brand the model as a tool to be used by their customers, will be deemed a GPAI provider. Mere users (deployers) of third-party models do not fall within this definition.
A General Purpose AI model is defined by the AI Act as an AI model that:
- Displays significant generality, meaning that it is capable of competently performing a wide range of distinct tasks.
- Can be integrated into a variety of downstream systems or applications.
Typical examples of GPAI models include chatbots used for customer support and tools that draft text or generate images on request.
The AI Act imposes specific obligations on GPAI providers, with additional, stricter requirements for GPAI models deemed to pose a “systemic risk” due to their advanced capabilities or widespread impact.
Purpose of the Code of Practice
The Code is designed to help industry stakeholders comply with AI Act rules concerning general-purpose AI models, particularly addressing systemic risks and transparency requirements. It was developed with input from nearly 1000 stakeholders and independent experts.
Voluntary nature and legal benefits
While voluntary, adherence to the Code allows GPAI model providers to demonstrate compliance with the AI Act, potentially reducing administrative burdens and providing clearer legal certainty compared to other compliance methods. Moreover, whilst signing up to the Code gives GPAI model providers the status of “good faith actors” and a 1-year grace period from regulatory repercussions (including fines) if they fail to fully implement their commitments, the same will not apply to non-signatories of the Code.
Models placed on the market before 2 August 2025 must comply with the AI Act obligations by 2 August 2027.
Structure of the Code
The Code is divided into three chapters covering Transparency, Copyright, and Safety and Security. The Transparency and Copyright chapters apply broadly to all GPAI providers, with the former chapter providing a tool to easily document the information necessary to comply with the AI Act transparency obligations, whilst the Copyright Chapter provides practical solutions for the implementation of a copyright policy. The Safety and Security chapter, on the other hand, targets providers of advanced models with systemic risks, outlining concrete practices for managing these risks.
Training data summary template
Alongside the Code, the AI Office is developing a detailed template for summarizing training data, which providers must publicly disclose. This template is closely linked to transparency and copyright obligations and has been shaped through extensive stakeholder consultations.
Next steps
In the coming weeks (up to 2 August 2025), Member States (through the AI Board) and the Commission (through the AI Office) will assess the Code of Practice’s adequacy. Additionally, the code will be complemented by Commission guidelines on key concepts related to general-purpose AI models.
From here on, GPAI model developers will have to ensure that they comply with the AI Act obligations and demonstrate their compliance. In so doing, they will need to decide whether to sign up to the Code of Practice, or otherwise find alternative ways of demonstrating compliance with their obligations under the AI Act.
