European Parliament adopts its negotiating position on the EU AI Act

The European Parliament fixed its negotiating mandate for the Artificial Intelligence Act on 14 June 2023, adopting amendments by 499 votes to 28 with 93 abstentions. The file then went back to committee for negotiations with the other EU institutions.

What it is

This was not law yet. It was the amended text Parliament carried into interinstitutional talks, and it shows where MEPs wanted the rules to go. Parliament's version also proposed an AI Office constituted as an independent Union body.

What it requires

Parliament's text lengthened the banned list. It covered untargeted scraping of facial images to build recognition databases, and emotion recognition in policing, border control, workplaces and schools. On the high-risk side, MEPs added security components used to run water, gas, heating, electricity and critical digital infrastructure. Generative systems built on general-purpose models would have had to disclose AI-generated output and publish detailed summaries of copyrighted training data. Research work and open-source components received exemptions.

What security leaders should do

Treat this page as history, not as the rulebook: compliance work should cite the adopted regulation rather than a negotiating mandate. The position is still useful for reading intent. Operators of utilities and critical digital infrastructure were singled out early, so security teams in those sectors should already know which AI components sit inside control and monitoring systems. When a vendor pitches a generative model, ask how generated content is disclosed and what training data is summarised, and keep that record with the procurement file.