The EU completely banned AI with an unacceptable risk

The first requirements of the “AI ACT) – a large -scale document regulating the use of artificial intelligence (AI) entered into force in the countries of the European Union. This law, approved by the European Parliament in March 2024, prohibits II systems representing an “unacceptable risk” for society. The law divides technology into four levels of risk-from the minimum (for example, spam filters) to unacceptable, which is now completely prohibited.

Image source: Alexander Lallemand / Unsplash

According to TechCrunch, the ban-systems used for social scoring (a person’s assessment according to his social characteristics), manipulation of the behavior, predicting crimes based on the appearance, analysis of people’s emotions at work or at school, as well as collecting biometric data in real time. In public places. Companies that violate these rules run the risk of receiving fines up to € 35 million or 7 % of the annual turnover, depending on which amount is greater. However, according to Rob Sumroy, the head of the technological direction in the Law Firm Slaughter and May, the fines will not be applied immediately: “Organizations should be completely ready to implement the rules on February 2, but the main measures will come into force only by the end of the summer “.

It is worth noting that some companies began to prepare for new requirements even before the law entered into force. So, in September 2024, more than 100 organizations, including Amazon, Google and Openai, signed a voluntary agreement on the immediate application of the AI ​​ACT principles. This document ordered them to take into account the provisions of the law in advance and evaluate the risks of their technologies. However, large players such as Meta✴ and Apple refused to participate in the pact. Nevertheless, as Samra notes, this does not free the company from fulfilling the requirements of the law.

The law also provides for exceptions to a number of prohibitions. For example, law enforcement agencies can use biometric systems in public places to search for missing people or prevent a threat to life if permission of the relevant authorities is obtained. It is also allowed to use systems that analyze emotions in schools or at workplaces, if it is justified by medical or therapeutic purposes.

At the same time, experts warn of possible difficulties associated with the “pads” of the new law with existing norms such as GDPR, NIS2 and Dora. “The laws on the regulation of AI do not exist separately on their own, in a vacuum,” emphasizes Samra. “Companies will have to take into account the interaction of various legal acts with each other, especially in terms of notifications of violations and incidents.” The full implementation of all provisions of the law, including fines, is expected by August 2025.