A wind of panic is blowing over the AI Act. Negotiations around the draft European regulation on artificial intelligence (AI) have been very tense since MEPs slammed the door on a negotiation meeting with representatives of member states on Friday November 10, reported the EURACTIV agency.
The reason for their wrath? The elected officials had just learned that France and Germany, supported by Italy, were opposed to a compromise route on the “foundation models” : the three countries no longer wanted to regulate these artificial intelligence systems which allow the creation of text or images, and which can be used by companies in applications such as the ChatGPT or Bard conversational robots.
Disagreements are common in “trilogues” – the negotiation of a final text between the European Parliament and the Council of Member States, under the leadership of the Commission. But this blockage could call into question the agreement on the text, subject to a timetable emergency. And France’s attitude arouses criticism.
According to our information, in an attempt to find a way out of the crisis, Paris will, through the Minister Delegate in charge of digital technology, Jean-Noël Barrot, propose to apply to foundation models a base of minimum transparency obligations, such as the publication of a summary of the characteristics of the software as well as its operation and of the way he was trained. These rules could be brought together in a code of conduct adopted by companies. This approach was to be submitted to the other member states during the next Council meetings, Friday 17 and Tuesday 21 November.
“The EU risks wasting a historic opportunity”
With this offensive, France, Germany and Italy oppose the obligations planned for the spring by Parliament. The latter called for regular assessments and mitigation measures of the risks posed by these models for the security and fundamental rights of the European Union (EU), which would include combating errors, illegal content, bias discriminatory… These countries consider these procedures too cumbersome and too costly for European start-ups. They also reject the compromise put forward at the beginning of November, which provided for only applying these obligations to the most powerful models.
To ward off criticism, Paris claims not to be opposed to the regulation of AI and believes that the AI Act would already be the most restrictive text in the world. France argues that foundation models will be subject to strong obligations if they are used by companies for purposes “high risk”, which constitute the initial basis of the AI Act: autonomous driving, exam scoring, CV sorting, health diagnosis, obtaining credit, use in justice, etc.
You have 50% of this article left to read. The rest is reserved for subscribers.
Gn Fr headline