On July 30, 2024, the European Commission announced the launch of a consultation on trustworthy general-purpose artificial intelligence (“GPAI”) models and an invitation to stakeholders to express their interest in participating in the drawing up of the first GPAI Code of Practice (the “Code”) under the newly passed EU AI Act (see our previous blog here). Once finalized, GPAI model providers will be able to voluntarily rely on the Code to demonstrate their compliance with certain obligations in the AI Act.

Consultation

The consultation provides stakeholders with the opportunity to have their say on topics that will be covered by the Code. It will also inform the AI Office’s development of the template summary of training material that GPAI model providers will be required to publish under Article 53(1)(d) of the AI Act.

The consultation covers three topics:

  1. Transparency and copyright: This relates to the documentation and policies that providers of GPAI models should have in place to comply with EU copyright law. Part D relates to the content and level of granularity expected from the template summary of training material referenced above.
  2. GPAI models with systemic risk: This relates to how the systemic risks associated with certain GPAI models should be classified, identified and assessed, mitigated, and internally governed (through policies and procedures).
  3. Reviewing and monitoring the GPAI Code of Practice: This relates to how the AI Office will encourage and facilitate the review and adaptation of the Code after its initial implementation.

Interested parties can submit their responses to the consultation via an online form by September 18, 2024. They also have the option to share additional information with the AI Office by filling out the template document featured at the end of the questionnaire.

Call for Expression of Interest

The Commission intends to finalize the Code by April 2025, i.e., 9 months from the AI Act’s entry into force on August 1, 2024. Stakeholders can express their interest in participating in the drawing up of this Code via this application form by August 25, 2024.

Once stakeholders have expressed this interest, the AI Office will select a number of them, including model providers, civil society organizations, academics, and independent experts, to be part of a “Code of Practice Plenary.” The first of these plenaries will be convened in September 2024.

Stakeholders will then collaborate in one or more of four Working Groups, to address specific issues such as risk mitigation, internal risk management, transparency and copyright, and risk identification and assessment. Plenary participants will also be asked for their feedback on the template summary of training material referenced above.

As the main addressees of the Code, GPAI model providers will also be invited to dedicated workshops during this time, in addition to their plenary participation.

* * *

The Covington team continues to monitor and advise on AI developments in the EU, UK, US and beyond. Please get in touch with a member of the team if you have any questions.

This blog was drafted with the contribution of Salena Mann.