One of the points of the new rules is transparency. The creators of the generative models will be responsible for disclosing information about the materials used for training, as well as marking the content as created by artificial intelligence. Limiting the work of artificial intelligence for mass biometric surveillance and smart police activities has also become an important aspect.

what is known

The following mechanisms of biometric AI systems are banned:

• Remote biometric identification systems in real time in public places.
• “Delayed” remote biometric identification systems – for the purpose of criminal prosecution for serious crimes and only with court enforcement, excluding law enforcement systems.
• Biometric classification systems using key characteristics (gender, race, ethnicity, citizenship status, religion, political beliefs).
• Predictive policing systems (based on profiling, location analysis and past crimes).
• Emotion recognition systems in law enforcement, border services, workplaces and educational institutions.
• Indiscriminate collection of biometric data on social networks or security cameras (violates the right to privacy).

The law also introduces the concept of “high risk” artificial intelligence systems that could potentially harm a person. Among them are the recommendation algorithms of social networks. Its developers will need to provide high-quality data for AI training and register the systems in a European database.

When using chatbots, developers need to inform users that they are interacting with artificial intelligence and not a live operator. For generative neural networks, there is also a requirement to specify the authorship of the content. So it should be clearly stated that the image or text is produced by AI.

The new rules also include obligations regarding environmental impact assessment.