Rapporteurs from the European Parliament’s different political groups continued their work at a meeting on Tuesday, 30 August to reach a compromise on legislation on artificial intelligence (AI) (EUROPE 12996/6).
Several relatively minor changes, compared to the previous version of the text, had been proposed in advance of the meeting. Thus, the new compromise foresees, among other things, that users of high-risk AI systems will have to cooperate with national competent authorities and with the future European AI Council to demonstrate the compliance of an AI system. The original text provided that only suppliers would be required to do so.
In addition, a new reference has been added to explicitly state that suppliers must inform without delay the relevant actors - primarily distributors - as well as the competent bodies and national supervisory authorities in case of non-compliance or corrective measures.
In addition, AI system providers are now required to keep logs automatically generated by their own systems for a period of at least six months. However, longer periods could be decreed depending on the sector in which the system is active.
These logs should be accessible to the competent authorities at EU level as well as to national market surveillance authorities upon reasoned request.
In addition, some of the discussions focused on administrative procedures and the sharing of responsibilities. In this respect, the text proposes, among other things, that suppliers of AI systems must appoint a legal representative responsible for ensuring that the conformity assessment procedure has been carried out; otherwise, the system cannot be placed on the European market. (Original version in French by Thomas Mangin)