The European Union’s AI law, a risk-based plan to regulate artificial intelligence applications, has passed what appears to be the last major hurdle standing in the way of approval after member state representatives voted today to confirmation of the final text of the draft law.
The development follows the political deal reached in December — concluded after marathon “final” tripartite talks between EU co-legislators lasting several days. After that, work began to turn the agreed positions on the junk negotiating sheets into a final compromise text for lawmakers to approve — culminating in today’s EMA vote confirming the draft rules.
The planned regulation sets out a list of prohibited uses of AI (aka unacceptable risk), such as using AI for social scoring; introduces certain governance rules for high-risk uses (where AI applications may harm health, safety , fundamental rights, the environment, democracy and the rule of law) and for the more robust general purpose/fundamental models deemed to pose ‘systemic risk’. and applies transparency requirements to applications such as AI chatbots. However, “low-risk” applications of artificial intelligence will not fall under the scope of the law.
The vote confirming the final text will lead to a huge sign of relief in much of Brussels. Continued opposition to risk-based AI regulation, led by France – fueled by a desire to avoid legal boundaries that prevent domestic AI startups like Mistral AI from turning into national champions that could challenge the rise of American AI giants. the potential for the regulation to be derailed, even at this late stage.
At the event, all 27 EU member state ambassadors unanimously supported the text.
If the vote failed, there was a risk that the entire regulation would be grounded, with limited time for any renegotiations – given the looming European elections and the end of the current Commission’s term later this year.
As for the approval of the bill, the baton now passes to the European Parliament, where lawmakers, in committee and in plenary, will also take a final vote on the compromise text. But given that the biggest backlash came from a few member states (Germany and Italy were also linked to doubts about the AI law that places obligations on so-called fundamental models), these upcoming votes seem academic. And the EU’s flagship AI law should be passed into law in the coming months.
Once approved, the law will enter into force 20 days after it is published in the EU’s Official Journal. There will then be a staggered implementation period before the new rules for AI applications and models come into force – with a six-month grace period before they take effect a list of prohibited uses of AI set out in the regulation (probably around autumn);
The phase-in also allows a year before rules for foundational models (also known as general-purpose AI) apply — so not until 2025. Most of the remaining rules won’t take effect until two years after the law is published .
The Commission has already moved to begin creating an AI Office that will oversee compliance of a subset of more robust fundamental models deemed to pose systemic risk. It also recently announced a package of measures intended to boost the prospects of domestic AI developers, including revamping the bloc’s supercomputer network to support the training of AI models.
