European Parliament adopts AI Act — the world’s first rules for artificial intelligence

The European Parliament has adopted the AI Act, which sets rules and requirements for developers of artificial intelligence models. The document aims to ensure the safety, transparency, environmental friendliness, and ethical use of AI in Europe.

The rules establish a risk-based approach to AI and formulate obligations for developers and users of AI depending on the level of risk it may pose. AI systems with an unacceptable level of risk to human safety will be strictly prohibited. This group will include, for example, systems that use subliminal or targeted manipulative methods, exploit human vulnerabilities, or are used for social assessment (classification of people based on their behavior, socioeconomic status, personal characteristics).

The Parliament also clarified the list of practices prohibited for artificial intelligence. It includes:

— remote biometric identification, in particular in public places in real time, with an exception for law enforcement agencies and only after a court order;

— biometric categorization using sensitive characteristics (e.g., gender, race, ethnicity, citizenship, religion, political position)

— predictive policing systems (based on profiling, location, or past criminal behavior);

— emotion recognition in law enforcement, border services, workplaces and educational institutions;

— indiscriminate extraction of biometric data from social networks or video recordings from CCTV cameras to create facial recognition databases (violation of human rights and the right to privacy).

The law clarifies the list of «high-risk» AI systems. According to the document, these are systems that can potentially harm human health or the environment. It also includes AI systems used to influence voters in political campaigns and social media recommendation algorithms.

Suppliers and developers of «high-risk» AI are required by law to conduct risk and compliance assessments, register their systems in the European AI database, ensure high quality data used for AI training, ensure system transparency and user awareness, as well as mandatory human supervision and the possibility of intervention in the system.

As for «low-risk» AI, such as chatbots, its providers are obliged to inform users that they are interacting with AI, not a human. For generative models, another requirement is added — to inform users that the content is created by a neural network.