European Commission publishes the General-Purpose AI Code of Practice

Background

The AI Act sets obligations for general-purpose AI models, and those rules enter into application on 2 August 2025. The Commission developed a code to help providers prepare for that date and to show how they can meet the relevant requirements in practice.^[1] General-purpose models can be used in many downstream applications, so the code is meant to give providers a common route to compliance rather than leave each company to interpret the rules alone.

Publication and drafting

The European Commission received the final version of the General-Purpose AI Code of Practice on 10 July 2025.^[1] The code was developed by 13 independent experts, with input from more than 1,000 stakeholders. Those stakeholders included model providers, small and medium-sized enterprises (SMEs), academics, AI safety experts, rightsholders and civil society organisations.^[1]

Structure of the code

The code has three chapters. The first two cover transparency and copyright and address all providers of general-purpose AI models. The third covers safety and security and is relevant only to a limited number of providers of the most advanced models.^[1] Within the transparency chapter, the code offers a Model Documentation Form that lets providers document the necessary information in a single place.^[1] In the copyright chapter, it offers providers practical solutions for putting in place a policy that complies with EU copyright law.^[1] This split means that every provider faces baseline transparency and copyright expectations, while the most advanced developers face additional safety and security provisions.

Voluntary participation and benefits

Providers can sign the code voluntarily. Those who do so can demonstrate compliance with the relevant obligations and benefit from reduced administrative burden and increased legal certainty.^[1] The code therefore works alongside the legal obligations rather than replacing them: signing is a way to show compliance, not a separate legal regime.

Enforcement and next steps

The Commission said the rules become enforceable by the AI Office one year later as regards new models and two years later as regards existing models.^[1] The code is also to be complemented by guidelines on general-purpose AI published ahead of the entry into force of the obligations.^[1] For providers, that means the code is one part of a wider compliance package: the AI Act sets the duties, the code gives a voluntary way to document and demonstrate compliance, and the guidelines are intended to add further guidance before the obligations become enforceable.

Why it matters

The code gives providers a single reference for documenting transparency, setting a copyright policy and, for the most advanced models, addressing safety and security.^[1] Because it is voluntary, it does not by itself create new legal duties; its importance lies in turning broad AI Act requirements into a more operational set of practices and in offering signatories reduced administrative burden and greater legal certainty.^[1] It also separates obligations that apply to all general-purpose AI providers from those aimed at the most advanced models.^[1]