Negotiators from the Council of the EU and the European Parliament met in the margins of the plenary session in Strasbourg on Monday and Tuesday, 2 and 3 October, to try to make progress in the interinstitutional negotiations (‘trilogues’) on artificial intelligence (‘AI Act’) (see EUROPE 13253/4).
During this third trilogue, co-legislators made progress on several points, notably approving the requirements for high-risk AI systems. They also agreed on the part of the text pertaining to fines in the event of failure to fulfil the obligations set out in the text as well as on the final provisions.
Furthermore, the negotiators from the Council of the EU and the European Parliament approved the point that suppliers of high-risk AI systems could be exempt from certain obligations after conducting a self-assessment.
Even so, this section is—at the European Parliament’s request—expected to be reworked at the technical level, as are a number of other points, such as the classification of AI systems designated as high-risk or the list of high-risk use cases included in Annex III.
Besides these issues, negotiators instructed the technical groups to work on the first two articles of the text, which pertain to general provisions and scope; on general-purpose AI systems and basic models (Articles 4a and 28b); on the articles pertaining to governance (Articles 56 to 59); and on the application of the legislation.
Approximately thirty articles had already been finalised before this third trilogue, which focused on, among other issues, notified bodies and even standards and assessments. The fourth round of interinstitutional negotiations will take place on 25 October. The Spanish Presidency of the Council of the EU plans to take stock [of the state of negotiations] with EU Member State permanent representatives (Coreper) on 6 October. (Original version in French by Thomas Mangin)