EU AI Act Negotiations Stall Over Foundation Models and Security Exemptions
European lawmakers reach a tentative agreement on regulating foundation models but fail to secure a final political deal on the AI Act due to ongoing disputes over national security exemptions and facial recognition bans.
EU negotiators hold the final round of trilogue discussions on the Artificial Intelligence Act but fail to reach a provisional political agreement due to significant differences between the European Parliament and EU governments. By the following morning, the involved parties manage to find a tentative compromise on foundation models, which includes a requirement for fundamental rights impact assessments for AI systems.
Despite this progress on foundation models, several major obstacles prevent a final deal. Officials continue to debate forbidden AI practices such as social scoring and the governance structures needed to oversee the law's implementation. A primary sticking point remains whether national security receives an exemption from the Act's scope, a demand pushed by EU governments but previously rejected by the European Parliament.
Additionally, EU governments strongly oppose the Parliament's proposed total ban on facial recognition technology, creating further division in the negotiation room. Reports indicate that at least one main political group pressures its members to align with the governments' position on these controversial issues, leaving industry observers like FEBIS to closely monitor the outcome of these delayed final negotiations.