The European Union’s member states today agreed on the text of a proposed law designed to regulate artificial intelligence models.

The legislation was first proposed in 2021 by the European Commission, the EU’s executive branch. Lawmakers reached a provisional agreement on what the law should include this past December. Now that the draft has received the greenlight from EU member states, it can proceed to a vote before the European Parliament, the last step involved in enacting a new law.

The proposed legislation is known as the AI Act. The version that received provisional approval in December called for a ban on harmful AI use cases such as emotion recognition in the workplace and educational institutions. Additionally, the draft contained rules designed to regulate the use of high-risk neural networks such as those used in the insurance and banking sectors.

The December version of the legislation would apply a second set of rules to so-called high-impact GPAI, or general AI, models. Those additional requirements specify that developers must check their models for “systemic risks” and mitigate any issues they find. Developers would also have to perform certain other tasks, such as reporting the energy efficiency of their neural networks.

The provisional approval that the AI Act received in December launched a weekslong negotiation process involving multiple EU member states. According to Politico, Germany and France hinted that they may oppose the text over concerns about its impact on AI innovation. Austria, in turn, is said to have expressed a similar stance because officials had qualms about the law’s privacy provisions.

To address those concerns, the European Commission reportedly introduced a “package of pro-innovation measures” for the AI sector. It also created a new regulatory body called the Artificial Intelligence Office that will be tasked with implementing the AI Act. 

An expert group comprising officials from EU member states will advise the European Commission on the implementation process. According to today’s report, one of the group’s priorities will be to ensure there’s no overlap between the AI Act and other regulations. The effort is expected to place a particular emphasis on a number of existing EU regulations that pertain to medical devices and machinery.

The AI Act is set to go before a committee vote in the European Parliament later this month. If it’s approved, the legislation will proceed to a full plenary vote. Politico reported that most of the European Parliament officials who worked on the AI Act expect the law will pass without any changes to the current version.

If it’s approved, the legislation will be rolled out in phases. The section of the AI Act that specifies which harmful AI use cases will be banned is reportedly set to go into effect six months after the law passes. Many of the legislation’s other provisions, including certain rules that pertain to high-impact GPAI models, will only apply from 2025 onward. 

Image: Unsplash