MEPs from the European Parliament’s Committee on the Internal Market and Consumer Protection (IMCO) and from the Committee on Civil Liberties, Justice and Home Affairs (LIBE) are preparing, on Thursday 11 May, after long months of internal negotiations, to vote on the report by Dragoș Tudorache (Renew Europe, Romanian) and Brando Benifei (S&D, Italian) on the Artificial Intelligence Act (‘AI Act’) (see EUROPE 13172/12).
“We can be proud of what we have done in just a few months. It will be important to get a strong majority at the plenary session in June to go into the interinstitutional negotiations with a strong and clear mandate”, said Mr Benifei.
Several changes were made to the final version of the text agreed by the European Parliament’s political groups. It now includes a number of provisions relating to artificial intelligence (AI) foundation models, which can be adapted to a wide range of tasks and purposes, and to generative AI systems. These types of systems were not covered by the Commission’s original proposal.
Stricter obligations are foreseen for those who will develop these systems, including documenting and being transparent about the hardware used to train the algorithms.
Sanctions inspired by those provided for in the DSA-DMA package
The text also foresees fines for foundation model providers of up to €10 million or 2% of the annual worldwide turnover generated.
“We first examined the Commission’s proposal. We then looked at similar sanctions used in other legislation such as the DSA (see EUROPE 13176/33) or the DMA (see EUROPE 13173/37). We are trying to have something consistent with other enforcement mechanisms for existing digital instruments and to equalise the impact of certain violations with similar violations under the DSA and the DMA”, Mr Tudorache summarised for EUROPE.
Regarding systems classified as ‘high risk’, MEPs agreed that an AI system must pose a ‘significant’ risk to be considered ‘high risk’. This risk will be measured through different criteria such as the potential severity, duration or the chances of it resulting in an incident. The recommendation systems of very large online platforms - as defined in the Digital Services Act - will fall into the category of high risk systems.
Maintaining a risk-based approach, a provision has also been added to oblige suppliers to carry out a fundamental rights impact assessment.
Separate vote on biometric identification
A separate vote will also be held on the issue of banning biometric identification, which the EPP opposes.
“This will be a separate vote on a point that is not part of the compromise. On the text, we should have a large majority and there should not be many changes. Perhaps on aspects such as fundamental rights, where some MEPs may want to advance their views before the plenary vote, but we have strong support with this compromise. I don’t see any big changes taking place”, Mr Benifei said.
The creation of the future Office for Artificial Intelligence is also confirmed in the text. This entity will be responsible for supporting the implementation of future legislation.
“In my opinion, it will take one to two years before the AI Act is fully effective. This period will be used at national level to allow governments to prepare their national policy, hire staff and prepare for its implementation. The same is true at European level. This period of one to two years will therefore also serve to prepare the Office for AI at European level”, Dragoș Tudorache explained to EUROPE. (Original version in French by Thomas Mangin)