The European Union’s Artificial Intelligence Law, a risk-based plan to regulate AI applications, has passed what appears to be the final big hurdle standing in the way of its adoption after member state representatives voted today to confirm the final text of the bill.
This development comes on the heels of the political agreement reached in December – which was reached after marathon three-way “final” talks between EU co-legislators that extended over several days. Then, work began to transform the agreed-upon positions on the pessimistic negotiating papers into a final settlement text for lawmakers’ approval — culminating in today’s Coreper vote to confirm the draft rules.
The planned regulation sets out a list of prohibited uses of AI (also known as unacceptable risks), such as using AI for social recording; Brings some governance rules for high-risk uses (where AI applications may harm health, safety, fundamental rights, environment, democracy and rule of law) and for the strongest general purpose/basic models deemed to pose a “systemic risk”; It applies transparency requirements to applications such as AI-powered chatbots. But “low-risk” AI applications would not fall within the scope of the law.
A vote on the final text will result in a huge sign of relief in most parts of Brussels. Continued opposition to risk-based AI regulation, led by France — and fueled by a desire to avoid legal limits that stand in the way of scaling home-grown AI startups like Mistral AI into national champions who might challenge the rise of US AI giants — has threatened . The possibility of the organization derailing, even at this late stage.
At the event, all 27 EU Member State Ambassadors gave the text their unanimous support.
Had the vote failed, there would have been a risk of the entire regulation collapsing, with time limited for any renegotiations – given the looming European elections and the expiration of the current Commission’s mandate later this year.
Regarding the adoption of the draft law, the baton now returns to the European Parliament where lawmakers, in committee and plenary, will have a final vote on the text of the settlement. But given the greater backlash that has come from a handful of member states (Germany and Italy have also been linked with doubts about an AI law that imposes obligations on so-called basic models), these upcoming votes seem academic. The EU’s groundbreaking AI law is set to be adopted into law in the coming months.
Once adopted, the law will enter into force 20 days after its publication in the Official Journal of the European Union. There will then be a staggered implementation period before the new rules apply to in-scope applications and AI models – with a six-month grace period before the list of prohibited uses of AI set out in the Regulation comes into effect (likely around the autumn).
The phased entry into force also allows a year before the rules apply to foundational models (also known as general-purpose AI) — so not until 2025. The bulk of the rest of the rules won’t come into effect until two years after the law is published.
The Commission has already moved to begin establishing an Office for Artificial Intelligence that will oversee compliance with a subset of the more robust foundational models deemed to pose systemic risks. It also recently announced a package of measures aimed at boosting the prospects of local AI developers, including retooling the EU’s supercomputer network to support training on generative AI models.