On 8 December, the European trilogue (EU Council of Ministers, EU Commission and EU Parliament) agreed on the main lines of the European regulation of artificial intelligence. This regulation is the culmination of the European strategy on the development of our information society and is justified by the desire for a sovereign Europe, both in the protection of its values and its economic interests. The contribution aims to show how the AI regulation illustrates a new regulatory approach, already present in the other texts but not in such a way. After these remarks, the contribution intends to analyse at the light of the final compromise but also by referring to the previous versions of the text the main provisions of the AI Act. How has the text finally integrated the “foundation models?” How has the text enlarged the risks to consider by the main actors and following which values they have to assess and if needed to mitigate these risks? Which categories of AI systems, the AI Act is distinguishing according to the gravity of the risks generated and which obligations are linked to each of these categories? Which governance, including as regards the standardisation procedures, has been put into place to control the respect of these obligations and the future developments of the AI systems? Finally, how the text is balancing the principle of precaution with the concern of innovation?