EU Member States will work on the third version of the Czech Presidency’s compromise text for the second part of the proposal on harmonised rules on artificial intelligence (AI) (see EUROPE 13023/26) in a meeting of the EU Council’s telecommunications working party on Thursday 29 September.
In detail, the Czech Presidency proposes that, with regard to the specific characteristics of general purpose AI systems and the requirements for suppliers, the European Commission should be required to adopt an implementing act specifying how the requirements are to be applied, based on a detailed impact assessment, rather than applying certain requirements directly. This analysis should also take into account technical feasibility and market and technological developments.
Still on the subject of suppliers’ obligations for general purpose AI systems, several clarifications have been made. Thus, wording deemed “clearer” has been added, with the aim of ensuring that AI system providers take all “necessary and proportionate” measures to prevent potential new misuse.
The compromise document also looks at high-risk AI systems. In this area, several clarifications have been made concerning standards, conformity assessment and registration certificates.
On this point, the Czech Presidency wants to limit the Commission’s discretionary power with regard to the adoption of implementing acts laying down common technical specifications for the requirements for high-risk AI systems and general purpose AI systems.
The Commission could adopt these acts after consultation with the ‘European Council for AI’: - where no reference to harmonised standards “covering the relevant essential safety or fundamental right concerns is published in the Official Journal of the European Union”; - when it has asked one or more European standardisation bodies to draw up a harmonised standard; - where the request has not been accepted by one of the European standardisation organisations; - when the standard is not delivered within the deadline.
In addition, public authorities, agencies or bodies using high-risk systems would be required to register in the EU database of high-risk systems. Suppliers should register themselves and their system in the database.
Concerning the ‘European Council for AI’, the text has been completed to include the obligation to create a platform for cooperation between market surveillance authorities and notifying authorities. The European Commission could carry out market assessments in support of the activities of the permanent market surveillance sub-group.
See the document: https://aeur.eu/f/39l (Original version in French by Thomas Mangin)