The European Parliament’s co-rapporteurs and shadow rapporteurs on the ‘omnibus’ simplification of legislation on artificial intelligence (AI Act) are nearing the end of their negotiations. In their latest compromise amendment proposal, dated Wednesday 11 March and available to Agence Europe, the elected representatives propose several changes to the initial text, and add a ban on AI models that generate sexual deepfakes, following the example of the Council of the EU (see EUROPE 13807/5).
The co-rapporteurs propose to keep the timetable for the date of entry into force of the obligations for high-risk AI models, as did the Council (see EUROPE 13822/13), but delete the notion of “strict” necessity for the possibility for non-high-risk AI models to use special categories of personal data to detect bias.
This particular point has been the subject of debate, with the first draft of the report having amended this possibility to be applied “by way of exception” only.
In addition, like the Council of the EU (see EUROPE 13825/5), the co-rapporteurs add a subparagraph to Article 5, designed to prohibit “placing on the market, putting into service or using an AI system that artificially modifies, manipulates or generates realistic images or videos in order to depict sexually explicit activities or the private parts of an identifiable natural person without that person’s consent”.
This ban would not apply to systems that have put in place “effective security measures to prevent the creation of such representations”, the amendments state.
Finally, they propose to amend the Annexes, in particular sections A and B of Annex I, so that sectoral legislation takes precedence over the AI Act and, in theory, reduces the risk of legislative overlap.
A final technical meeting is scheduled for 13 March, to finalise the outstanding points. The text is due to be voted on in committee on 18 March. (Original version in French by Isalia Stieffatre)