EU negotiators strike a political deal on the AI Act
Foundation models and police use of biometric surveillance were the last sticking points; the deal set tiered duties for both, with the text finalised in 2024.
- Government & policy
- Era-defining
After three days of negotiation, running through the night of 8 December 2023, representatives of the European Parliament and the Council of the EU reached a provisional political agreement on the Artificial Intelligence Act. The deal, announced early on 9 December, unblocked a law that had been in draft since the Commission’s 2021 proposal and settled the two questions that had held it up.
The Act takes a risk-based approach. It bans a small set of uses judged to pose an unacceptable risk, imposes detailed obligations on systems classed as high-risk, and applies lighter transparency duties to others. The first sticking point was how to treat foundation models — general-purpose systems such as the large language models released over the previous year, which do not fit neatly into a use-based scheme. Negotiators agreed to add obligations aimed at these models directly, including technical documentation, summaries of training data intended to help rights-holders check for copyrighted material, and stricter requirements for the most capable models judged to carry systemic risk. Some member states, France and Germany among them, had pressed for lighter treatment to avoid disadvantaging European developers.
The second was biometric surveillance. Parliament had wanted a full ban on real-time remote biometric identification in public — face recognition at scale — while member states wanted exemptions for law enforcement. The compromise permitted narrow police use under judicial authorisation for defined serious cases, with safeguards that critics said were still too broad.
A political agreement is not a finished law. Technical drafting continued into 2024; the Parliament and Council formally adopted the text in the following months, and it entered into force in August 2024, with its provisions phasing in over the following years. The December deal is the point at which the EU committed to binding, horizontal AI rules — the first jurisdiction to do so — and set a reference point that other governments measured their own approaches against.