The Member States will meet again on Thursday 23 November, at a meeting of the EU Council’s Telecommunications Group, to try to agree on a common approach to foundation models as part of the interinstitutional negotiations (‘trilogues’) on the artificial intelligence (AI) legislation (see EUROPE 13296/5).
On this occasion, they will try to find a solution to break the deadlock after France, Germany and Italy opposed the Spanish Presidency of the EU Council’s tiered approach to the issue (see EUROPE 13291/9), which had proposed opting for stricter criteria and rules for ‘high-impact’ foundation models (see EUROPE 13279/18).
Among other things, the three Member States would like foundation models not to be governed by rules set out in the text, but by a code of conduct, and for sanctions not to be imposed in the first instance.
On the European Parliament side, the negotiators would like the foundation models to comply with the rules that will form part of the final text. In an attempt to reach a compromise, at the beginning of the week the Commission paved the way for a reduction in obligations and the introduction of codes of conduct, while maintaining the tiered approach.
“We cannot accept the proposal from Germany, France and Italy on foundation models. Even the minimum self-regulation standards should cover transparency, cybersecurity and disclosure requirements – which is exactly what we are calling for in the AI Act. We cannot turn a blind eye to the risks”, commented Axel Voss (EPP, German), the rapporteur for the future legislation on behalf of the European Parliament’s Committee on Legal Affairs.
The MEPs involved in the negotiations with the EU Council on AI continued their work on several aspects on Tuesday 21 November. Discussions focused in particular on the role of the future European Council for Artificial Intelligence.
This organisation could be tasked with overseeing the implementation of the regulation, carrying out assessments to monitor the compliance of AI systems for general use, or conducting investigations. The European Commission has proposed that this body should consist of one representative per Member State, appointed for a three-year term. This appointment could be renewed once.
MEPs would also like the European Data Protection Supervisor, the Fundamental Rights Agency and the European Union Agency for Cybersecurity (ENISA) to play a role. (Original version in French by Thomas Mangin)