The EU27 Ambassadors to the EU, meeting in the Committee of Permanent Representatives on Friday 13 March (Coreper), adopted their common position on the revision of the AI Act, which is part of the ‘omnibus’ package for simplifying digital legislation presented by the European Commission last November (see EUROPE 13755/3).
The approved text, which we have already detailed (see EUROPE 13822/13), puts the brakes on the Commission’s ambitions and redefines several key aspects of the proposal: prerogatives of the AI Office, fixed deadlines, controlled use of personal data by AI models, safeguards for real-life tests.
The last-minute addition to Article 5 on the ban on AI systems capable of generating non-consensual sexual ‘deepfakes’ and child sexual abuse material has been retained (see EUROPE 13825/5).
According to our information, however, Sweden has requested an amendment to this specific article, to allow national law enforcement agencies to use artificial intelligence to generate child sexual abuse material “for the purpose of conducting criminal proceedings or preventing, detecting or investigating criminal offences”.
This Article 5, introduced by a joint effort by France and Spain (see EUROPE 13807/6), has been the subject of several comments from member countries who feel that its wording could have been clarified. According to several sources, Austria, Hungary, Italy and Germany have expressed doubts.
Berlin has also tabled a number of amendments to Articles 9 and 11 to go further in terms of administrative simplification.
Now that the EU Council has adopted its position and negotiating mandate, it is up to the European Parliament to finalise its own position. The co-rapporteurs on the text are in the process of producing a final version, which should be voted on by the Committee on Civil Liberties (LIBE) and the Internal Market (IMCO) on Wednesday 18 March.
The elected representatives plan to hold a plenary vote the following week, on 25 March. Inter-institutional negotiations (trilogues) could therefore start at the beginning of April and be concluded before the date of entry into force of the rules for general-purpose AI models (GPAI) and the start of potential sanctions on 2 August. (Original version in French by Isalia Stieffatre)