The EU-27 unanimously adopt the AI Act

On February 2, 2024, the European Council unanimously approved the AI Act, the draft European regulation to regulate so-called « foundation » AI models. The European Parliament still has to approve the final version of the text, which should only be a formality. The AI Act should therefore be adopted during this term. However, threats of vetoes by Germany, Austria and, above all, France had put this scenario in jeopardy.

These three countries had strongly criticized the version of the AI Act presented in December 2023. In particular, French President Emmanuel Macron had said it was « not a good idea » to regulate AI « much more » than the rest of the world. He feared that too many constraints would hold back Europe’s AI champions, notably France’s Mistral AI.

In the end, France did win a few concessions. One amendment grants an exception in the name of « business secrecy » to the transparency obligations on training data for generative AIs. Another promises a regular review of the AI Act’s obligations to ensure that they do not hinder the development of European leaders.

But for the most part, the text remains unchanged. It still plans to classify AI models according to three levels of risk, leading to three levels of regulatory constraints. It also retains specific obligations for generative AI solutions.

« The AI Act has caused quite a stir…and rightly so! Today, the 27 Member States have approved the political agreement reached in December – acknowledging the perfect balance struck by the negotiators between innovation and security, » said a delighted Thierry Breton, European Commissioner for the Internal Market.