AI Act: 5 questions to understand this new regulation on AI

© Vitor Miranda/Shutterstock

Last May, the AI ​​Act was adopted by the Member States of the European Union. This ambitious text aims above all to protect the rights of citizens and limit the misuse of AI. While the first provisions of this legislation came into force on Sunday, February 2, here is everything you need to know about this new system.

In detail, the text now prohibits all private or public social rating systems, such as those observed in China. Predictive policing devices that aim to estimate whether a person is more likely to commit crimes are also prohibited.

The same goes for applications of ” emotion recognition “ at work and school, details Le Monde. The identification of citizens by facial recognition in real time in public spaces and the biometric categorization of people to identify ” their race, political or religious opinions, sexual orientation or union membership “, are finally banned.

Our colleagues specify that from August 1, 2025, “general purpose” AI models, in other words text and image generators such as ChatGPT or Gemini will have to demonstrate transparency in their technical documentation and training data. Security audits on cyber risks will also have to be carried out.

Much later, in 2027, AI deemed to be high risk, such as those operating in water, electricity, road infrastructure, in justice, in the police or in education, will have to conduct assessments concerning the risks of errors and biases.

200% Deposit Bonus up to €3,000 180% First Deposit Bonus up to $20,000

To ensure compliance with the rules of the AI ​​Act, an AI Office will be created. It will work in conjunction with national data protection authorities, such as the National Commission for Information Technology and Civil Liberties (CNIL) in France. If they contravene the rules, the companies concerned are exposed to hefty fines of up to 7% of their global turnover.

As soon as it was adopted, the AI ​​Act was heavily criticized by players in the AI ​​sector. According to Le Monde, it is in particular the dissemination of a sufficiently detailed summary of the content used to train the large models that is problematic.

Rights holders welcome this principle, because they hope to receive compensation if their productions have been used. But the tech giants, on the contrary, oppose the argument of business secrecy to reject this obligation.

Some observers have argued that the EU could shoot itself in the foot by imposing restrictive standards on AI companies. French Tech is also complaining about the risk of regulatory layers that would slow down European innovation.

Quoted by Le Figaro, Kirsten Rulf, associate director at the Boston Consulting Group (BCG), co-author and negotiator of the EU AI Act, on the contrary defends the approach of the text:

The AI ​​Act provides an action plan to improve the quality of these systems and therefore, on the contrary, promote innovation. We are dealing with technologies that are still very unstable, which can make mistakes and jeopardize a company if the risks are not well controlled.

Mahasti Razavi, a partner at August Debouzy specializing in tech issues, echoed the same sentiment. Interviewed by Les Échos, the expert clarified: “To simplify, this text is there to create a framework of trust.”.

In the coming months, the AI ​​Act promises to be much talked about. And it could well be part of the discussions between the EU and the government of Donald Trump, which should support the grievances of the tech giants.

📍 To not miss any news from Presse-citron, follow us on Google News and WhatsApp.