European Parliament imposes strict rules on generative AI

The European Union is on track to become the first global player to adopt a governing framework for artificial intelligence (AI). The European Parliament communicated its position on Thursday in the run-up to negotiations with member states and the European Commission. The draft of the AI Act has added additional obligations for generative AI, which can create texts, images and other content.

"This is the first legislation of its kind and allows the EU to lead the way in making AI humane, reliable and safe," said Romanian liberal Dragos Tudorache, who is helping to steer the AI law. He claims this is "probably the most important ruling of this legislature".

In the merged Internal Market and Civil Liberties committees, Parliament's negotiating mandate was approved on Thursday by 84 votes to 7, with 12 abstentions. AI systems will be classified according to their security risk. If that risk is unacceptably high, the application in question will be banned. One such example is the real-time use of facial recognition in public places and social scoring based on the Chinese model.

"Consumers can breathe a sigh of relief after the disappointing proposal from the Commission and the position taken by Member States so far"

Unlike the EU member states, which determined their position in December, when ChatGPT and DALL-E were discussed, the European Parliament wants to add separate rules to the law for generative AI systems. These will be "subject to additional obligations in terms of transparency, data quality and human supervision", says Tom Vandenkendelaere (CD&V). For example, it must always be indicated when content is AI-generated, no illegal content can be produced and summaries of copyrighted content used to "train" the system will not be allowed to be published.

The companies behind the generative AI applications can be fined up to 10 million euros or 2 per cent of their annual turnover if they fail to comply with EU legal requirements.

Nevertheless, the MEPs want to continue to stimulate innovation in the AI sector, for instance, by allowing exceptions to the legislation for generative AI to develop the models further. This would require regulatory AI testing environments where innovative systems can be created, tested and validated in a controlled environment.

'Ground-breaking legislation'

The European Consumer Organization (BEUC) is pleased with the vote in Parliament. "Consumers can breathe a sigh of relief after the disappointing proposal from the Commission and the position taken by Member States so far," said vice-director Ursula Pachl. "The fact that MPs want to give consumers specific individual rights is more than welcome. These include the right to be informed when they are the subject of a decision taken by a high-risk AI system, such as the application for a loan or an insurance policy, or the right to join a class action suit."

The rapid development of artificial intelligence is a particular point of attention in the drafting of European legislation. That is why the MEPs want AI to be defined as technology-neutral, ensuring future systems fall under the AI Act. "We are about to create groundbreaking legislation that must be able to withstand the challenges of the times," said Italian social democrat Brando Benifei, who is jointly responsible for the file.

Next month, the plenary session of the European Parliament will also vote on the AI law giving the mandate the broadest possible support. Following this meeting, negotiations with the Member States and the Commission will start. Spain, which presides over the European Councils of Ministers from July to December, wants to reach an agreement before the end of the year.

^{© LIONEL BONAVENTURE / AFP}