Generative Artificial Intelligence: Legal Regime for Learning and Using Created Objects

The main legal problems arising in the use of generative artificial intelligence (hereinafter - AI) are the use of protected objects and data for training AI without appropriate consent, the impossibility of protection of generated objects and the ability to mislead in the absence of indication of artificial origin. To solve them, it is necessary to ensure the legality of data collection for training, determine the potential legal status of generated objects and define the rules for their use. The paper studies and analyses the regulation of the Russian Federation, the USA, the EU and the People's Republic of China in this topic and the relevant institutions for controlling and ensuring the legality of AI training, criteria for the protectability of generated objects. The possibility of applying, in case of development of requirements for AI training in the Russian Federation, checklists with requirements and development of a system of warnings in case of their violation was determined. The institute of publishing summaries of the used data and the system of assessing the protectability of such objects from the USA can also be applied. In the analysis of the ways to ensure the legality of the use of personal data for training, it was concluded that it is necessary to apply consent as a basis for processing with the allocation of criteria, the introduction of the concept of anonymisation was proposed, and for the results of intellectual activity it is necessary to consider the possibility of identifying a special case of open licence with special provisions, as well as the application of the doctrine of fair use. The paper analyses the possible legal statuses of the generated objects and identifies the currently applicable legal constructions. It is proposed to fix in the terms of use of services certain provisions protecting companies from liability for generating prohibited or protected content, shifting the obligation to verify the results and the corresponding risks to the user, but taking into account compliance with technical measures.