EU Advances AI Act: Draft of the General-Purpose AI Conduct Guidelines Now Available
The European Commission has introduced a voluntary compliance mechanism, the General-Purpose AI (GPAI) Code of Practice, to assist providers of general-purpose AI models in meeting their obligations under the AI Act. The Code, published on 10 July 2025, serves as a non-binding soft law instrument offering early guidance to AI developers before the GPAI provisions become legally enforceable from 2 August 2025.
### Scope and Content
Applicable to providers placing general-purpose AI models on the EU market, the Code focuses on advanced foundation models like ChatGPT and Google’s Gemini. It contains three main chapters: Transparency, Copyright, and Safety & Security. The Transparency and Copyright chapters offer a path for all providers to demonstrate compliance, while the Safety & Security provisions focus on a narrower subset of more advanced and potentially riskier AI models.
### Obligations and Requirements for AI Providers
Under the Transparency chapter, providers must ensure that their AI models disclose adequate information about their capabilities, limitations, and intended uses. The Copyright chapter requires providers to comply with intellectual property rights, ensuring lawful use and respect of copyright when training or deploying AI models. For models presenting systemic risks, enhanced safety and security measures must be implemented in accordance with the AI Act.
### Voluntary Nature and Benefits
The Code is voluntary, and AI providers can choose to sign it by completing a signatory form and submitting it to the European Commission’s AI Office. Signing the Code allows providers to demonstrate compliance with AI Act obligations more straightforwardly, lowering administrative burdens and increasing legal certainty compared to proving compliance via other methods.
Once endorsed by EU Member States and the Commission, adherence to the Code offers a clear, collaborative, and legally supportive pathway to meet AI Act requirements, especially beneficial ahead of the enforceability dates for new and existing AI models set for one and two years after August 2025.
### Next Steps
The adequacy of the Code is currently being assessed by the European Commission and Member States. Complementary Commission guidelines on key GPAI concepts are scheduled for publication soon after the Code’s release to further support implementation.
Notably, major US and EU companies building popular GPAI models have already agreed to sign the Code. However, questions remain about the duration and scope of a potential grace period for the Code's signatories and the selective commitment of providers to parts of the Code.
In summary, the GPAI Code of Practice is a pioneering voluntary compliance tool that guides AI providers on transparency, safety, security, and copyright to meet the AI Act’s regulatory requirements, reducing compliance complexity and fostering safer, more transparent deployment of advanced AI models in Europe.
- The Copyright chapter within the GPAI Code of Practice requires AI providers to comply with copyright law, ensuring they respect intellectual property rights when training or deploying their AI models, promoting fairness in education-and-self-development and technology sectors.
- As part of the general-news, it has been reported that major US and EU companies building popular general-purpose AI models have agreed to sign the GPAI Code of Practice, demonstrating a commitment to voluntary adherence to self-regulating guidelines on transparency, safety, security, and copyright, aiming to facilitate smoother finance and implementation of advanced models in the European Union.
