ARTIFICIAL INTELLIGENCE: THE EU CODE OF CONDUCT FOR THE AI ACT IS HERE – WHAT DOES IT ENTAIL?

The Code of Conduct, a voluntary instrument developed by independent experts with input from over a thousand stakeholders, aims to help the industry comply with the rules on general-purpose AI models, which will be applicable starting August 2, 2025.

The European Commission has received the final version of the Code of Practice for Generative AI, which « will help companies comply with the obligations set out in the AI Act regarding safety, transparency, and copyright, » especially for general-purpose AI models (GPAI). The Code, which now awaits approval from both the Commission and the twenty-seven member states, provides clarifications on a series of rules established by the AI Act. These rules will apply from August 2, 2025, to GPAI model providers, including those with systemic risks such as OpenAI's GPT-4, Google's Gemini, and xAI's Grok.

What It's About

The Code is structured into three chapters and is a voluntary instrument developed by 13 independent experts with contributions from over a thousand stakeholders. These include model providers, SMEs, academics, AI security experts, rights holders, and civil society organizations. Its objective is to help the industry comply with the GPAI model rules, which will be applicable from August 2, 2025. Implementation will fall under the purview of the AI Office, whose establishment was provided for in the AI Act. With the GPAI model rules, Brussels aims to ensure that such models, placed on the European market – including the most powerful and advanced ones – are safe and transparent.

The Three Chapters of the Code

As mentioned, the Code consists of three chapters: transparency and copyright, applicable to all GPAI models, and safety, concerning only a limited number of providers of the most advanced models. Brussels will not consider GPAI model providers who do not fully implement the commitments undertaken immediately after signing the Code as non-compliant. Instead, the principle of good faith in implementing commitments will prevail. The spirit is one of collaboration, with the Commission seeking ways to achieve full company compliance. The EU executive will have full compliance powers, including the ability to impose sanctions, starting from August 2, 2026.

Once the Code has been approved by the Member States and the Commission, GPAI model providers who voluntarily subscribe to the Code can demonstrate compliance with the obligations of the AI Act by adhering to the Code. This way, signatories of the Code will benefit from reduced administrative burden and greater legal certainty compared to providers who demonstrate compliance through other means.

Commission Guidelines

The Commission will complement the Code with guidelines, which will be published before the general-purpose AI obligations come into force. The guidelines will clarify who falls within and who falls outside the scope of the general-purpose AI rules set out in the AI Act.

Henna Virkkunen, Vice-President of the Commission for Technological Sovereignty, Security, and Democracy, stated: « Today's publication of the final version of the Code of Conduct for general-purpose AI marks an important step in making the most advanced AI models available in Europe not only innovative, but also safe and transparent. Designed in collaboration with AI stakeholders, the Code is aligned with their needs. » Virkkunen invited « all general-purpose AI model providers to adhere to the Code, » emphasizing that « this will ensure them a clear and collaborative path to compliance with the European AI Act. »