On Friday 2 February, the Permanent Representatives of the Member States to the EU (Coreper) validated the provisional political agreement on artificial intelligence (AI Act) reached by the European Parliament and the EU Council after more than 37 hours of negotiations on 8 December (see EUROPE 13311/17).
A source told EUROPE that there were few last-minute changes to the text. In recent weeks, however, several Member States, including France and Germany, have argued in favour of relaxing the requirements for certain aspects, such as the provisions relating to the most powerful AI systems, in order to protect their respective national champions.
On 30 January, Germany’s Minister for Digital Affairs, Volker Wissing, finally indicated that Berlin would support the agreement after the German Liberal Democrat Party (FDP) - a member of the coalition which had been arguing in favour of less stringent requirements under the AI Act - dropped its objections (see EUROPE 13340/13).
Although France was reluctant to support the text, it ultimately revised its position and followed Germany’s lead, after obtaining a regular update of the threshold above which AI may represent a systemic risk. In addition, on the subject of copyright, Paris also ensured that business secrecy was included in a recital of the text and that future delegated acts must take it into account.
A source also indicated that while there had been no change to the definition of foundation models representing a systemic risk, Paris would have to pay “close attention” in the drafting of the delegated acts.
“We feel that the economic ecosystem, businesses and citizens are all winners. The French position has become favourable because we have been heard”, explained Bercy.
A source close to the matter told EUROPE: “This is a unanimous agreement with an accompanying declaration that allows some of us to have our wishes met again and to have a text that’s balanced between rights and obligations, with rights and duties taken into account”.
Following validation of the political agreement in Coreper, France also announced that it would like to host the future European office for artificial intelligence. On 24 January, the European Commission adopted its decision to establish such an office, in accordance with the provisions of the AI Act (see EUROPE 13335/6).
“If part of the office is located in France, it will be welcome. (...) Wherever it is located, since it is a service and not an agency, we intend to be very closely involved in its work”, confirmed Bercy.
Italy had also expressed fears about the agreement, as had Austria. The latter also sent a statement to the other Member States expressing its concerns about data protection.
The text will now have to be voted on by MEPs before being formally ratified by the Member States. The vote in the Parliament committee should take place on 13 February, before a plenary vote in early April. (Original version in French by Thomas Mangin)