THE LIMITS OF AUTONOMY IN MAKING LEGALLY SIGNIFICANT DECISIONS BY ARTIFICIAL INTELLIGENCE SYSTEMS

The active integration of artificial intelligence systems into processes that produce legal consequences raises the question of the permissible limits of their autonomy. The aim of the study is to determine the legal criteria that constrain the autonomy of artificial intelligence systems when making legally significant decisions. The relevance of the topic stems from the fact that delegating an ever-wider range of decisions to algorithms while retaining the previous mechanisms of attribution creates a risk of weakening legal guarantees and blurring liability. The article successively examines the concept and features of a legally significant decision as applied to the operation of such systems; the relationship between the autonomy of artificial intelligence and the autonomy of will as a property of a subject of law; the gradations of autonomy and their legal significance; and the areas of private and public law in which artificial intelligence participates in making decisions that entail legal consequences. The methodological basis comprises the formal-legal, comparative legal, systemic-structural and functional methods. Drawing on Russian legal doctrine, strategicplanning instruments and foreign (supranational) experience of risk-based regulation, the author argues that the limit of autonomy is determined not by the nature of the technology but by the character and severity of the legal consequences of the decision, and that it should be fixed through the principle of mandatory preservation of meaningful human control. Criteria for distinguishing permissible from impermissible autonomy are proposed: the reversibility of consequences, the affecting of fundamental rights, the bound or discretionary nature of the decision, and its explainability and appealability. It is shown that bound, formalizable decisions allow a high degree of autonomy, whereas discretionary ones require human involvement. A three-tier model of the limits of autonomy is formulated, together with conclusions on the expediency of Russian regulation moving towards a riskbased model with the statutory consolidation of a mandatory human-oversight requirement for high-risk decisions. Funding. The study was carried out with the support of the Ministry of Science and Higher Education of the Russian Federation within the framework of the project “Fundamental problems of development methodology and the related legal and ethical regulation in the field of application of artificial intelligence systems and models” (FEWZ2024-0052)