On Tuesday 19 September, Brando Benifei, co-rapporteur for the legislation on artificial intelligence (AI) (see EUROPE 13232/10, 13201/1), updated the members of the European Parliament’s Committee on the Internal Market and Consumer Protection on the state of play in the interinstitutional negotiations (‘trilogues’). The next trilogue is scheduled for 2 and 3 October in Strasbourg, on the fringes of the Parliament’s plenary session.
So far, only two trilogues have been held. However, the first compromises are tending to emerge on a number of complex points, such as regulatory sandboxes, which allow suppliers to test their AI systems for a defined period without necessarily having to comply with the regulatory framework.
On this point, Italian rapporteur commented that regulatory sandboxes would be compulsory. However, Member States lacking the necessary resources could join forces with the regulatory sandboxes of another country or “share one”. This idea, which would satisfy the European Parliament, which wanted to keep the mandatory nature of the provision, and reassure the Member States, was one of the alternative options proposed by the European Parliament several months ago (see EUROPE 13219/12, 13109/8).
Discussions during the forthcoming trilogue will focus on a number of other important aspects, such as the question of the positive report on the exit from the regulatory sandboxes giving operators a presumption of compliance. For the time being, the EU Council is opposed to the proposed wording. No agreement was reached on AI testing in real-life conditions or on the assessment of fundamental rights, which the European Parliament has made one of its priorities.
Negotiations will also logically continue on the classification of high-risk systems. However, around thirty articles have been finalised so far, notably on notification bodies, standards and assessments, and certain obligations. (Original version in French by Thomas Mangin)