European Union member states unanimously agreed on Friday on the bloc's Artificial Intelligence Bill, overcoming last-minute concerns that the rulebook would stifle European innovation. After lengthy negotiations between Council representatives, European Parliament members and European Commission officials, representatives gave the green light to the final compromise text.
The law would ban certain applications of artificial intelligence technology, impose strict restrictions on use cases considered high-risk, and impose transparency and stress-testing obligations on state-of-the-art software models.
The EU is the first country to set binding rules for the rapidly developing artificial intelligence technology. While many countries and international clubs - from the OECD to the G7 - have spent the past few years thinking about how to regulate artificial intelligence, most have stuck with voluntary guidelines or codes of practice.
When EU policymakers announced in December that they had reached a final compromise on the content of the Artificial Intelligence Law, the breakthrough was hailed as a groundbreaking step that Europe should celebrate amid the rise of ubiquitous AI tools like OpenAI’s ChatGPT and Google’s Bard.
However, this result has caused dissatisfaction among some EU countries. Over the past few weeks, the group's top economies Germany and France, as well as Austria, have suggested they may oppose the text in Friday's vote.
Vienna's grievances lie over data protection provisions, while Paris and Berlin have warned that rules for advanced AI models will hinder the development of Europe's new AI champions, such as France's Mistral and Germany's AlephAlpha. With Italy (sometimes a critic of the AI Law) silent on its intentions, the fate of the AI Law was suddenly in question, as four opposing countries were enough to permanently derail the law.
The cabinet of French Economy Minister Bruno Le Maire called for a new round of negotiations with the European Parliament to resolve his concerns. The Belgian council president was alarmed by the prospect due to the lack of time for further negotiations. To make matters worse, the European Parliament itself is dealing with a debate over facial recognition rules in the Artificial Intelligence Bill, sparked by privacy hawk Svenja Hahn.
In the end, the problem was resolved through the EU's usual PR offensive and diplomacy. The European Commission has stepped up the pressure, announcing a package of major innovation measures in the field of artificial intelligence, and in one fell swoop established the EU Artificial Intelligence Office - an agency responsible for enforcing the Artificial Intelligence Law.
Two people involved in the negotiations said Austria, France and Germany were lured back to the negotiating table by promises that the committee would issue a formal statement offering reassurances on issues of concern to them. These statements are not legal acts, but since the Commission will oversee the implementation of the Artificial Intelligence Act, they will provide some kind of assurance.
A spokesman for German Digital Minister Volker Wissing, the leading skeptic of the Artificial Intelligence Act in the German coalition government, told the media: "We ask the European Commission to clarify that the Artificial Intelligence Act does not apply to artificial intelligence applications in medical devices."
A European Commission statement circulated by POLITICO among EU diplomats ahead of the vote showed plans to set up an "expert group" composed of authorities from EU member states. The function of the group will be to "advise and assist" the European Commission in the application and implementation of the Artificial Intelligence Law.
In particular, the expert group will assist the European Commission in avoiding duplication between the Artificial Intelligence Law and other EU regulations, including the Medical Devices Regulation, the In Vitro Diagnostic Devices Regulation and the Machinery Regulation.
The statement also instructs the Office of Artificial Intelligence to provide "detailed guidance" to developers of advanced "general-purpose" artificial intelligence models on how to disclose excerpts of copyrighted material used to train software. The statement also reiterated the Commission's commitment to promoting innovation in the field of artificial intelligence and "ensuring a flexible and future-proof legal framework" when updating certain texts affecting advanced artificial intelligence.
Finally, the declaration allows member states to adopt more restrictive rules and safeguards for technologies such as facial recognition, emotion recognition and biometric classification.
The Artificial Intelligence Act still requires formal approval by the European Parliament. The text will be endorsed at committee level in two weeks, with a plenary vote expected in April.
Disgruntled pro-privacy lawmakers may still try to block the law's progress by proposing amendments -- which, if passed, would require additional negotiations with the council. But most people involved in the drafting of the Artificial Intelligence Law in Parliament are confident that the law will be passed without any changes.