On Tuesday 22 February, EU Member States discussed the compromise text of the French Presidency of the EU Council (FPEU) for articles 40 to 52 of the proposed legislation on artificial intelligence (AI) (see EUROPE 12888/18).
In the chapter on harmonised standards, the compromise version of the FPEU specifies the general objectives – set by the Commission – to be achieved as a result of the standards that have been developed by the standards bodies.
The Commission’s standardisation requests will have to specify that the standards must be “consistent”, “easy to implement”, and they must ensure that AI systems respect EU values.
In addition, the text also provides for the Commission to be obliged to ask the standards bodies to provide evidence of the efforts they have made to meet the objectives of respecting EU values and strengthening global governance and cooperation in standardisation.
The compromise document now stipulates that where harmonised standards are deemed insufficient by the European Commission, the Commission may adopt common specifications for legal requirements for high-risk AI systems only after they have consulted with the European Artificial Intelligence Committee.
By way of reminder, the original proposal for harmonised AI rules foresaw the creation of this committee, intended to facilitate cooperation between national supervisory authorities and the Commission.
The European Commission would also be obliged to consult the European AI Committee before adopting delegated acts that would subject high-risk AI systems to the conformity assessment procedure as a result of technical progress that has been made.
In addition, the Commission would also be required to consult the Committee before adopting – as a result of technical progress made – delegated acts that update the EU Declaration of Conformity documents that have been drawn up by AI system suppliers.
These consultations with the committee, according to the paper, would ensure greater involvement of Member States in the delegated acts process.
In addition, the FPEU notes that the section of the compromise text covering AI systems placed on the market or in use by credit institutions, is still under review. Changes are likely to be made once the feedback from the European Central Bank has been incorporated.
With regard to law enforcement – a major issue for the European Parliament (see EUROPE 12876/18) – the FPEU amended Article 47 to add the possibility for law enforcement authorities to operate a specific high-risk AI system without authorisation in “duly justified” emergency situations.
This derogation would therefore introduce the possibility for the relevant authorities to make ex-post applications, provided that the application for authorisation is made “during or after use, and without undue delay”.
See the document: https://aeur.eu/f/i2 (Original version in French by Thomas Mangin)