Identification and practical interpretation of new features for classifying compliance risks in an artificial intelligence environment

In the context of the digital transformation of the compliance environment, not only are new relevant compliance tools emerging and developing, but a new category of «technogenic» risks is also being formed, for which traditional, static methods of systematization and grouping are of little use. The goal of this study is to identify new types of compliance risks from the use of artificial intelligence (AI) and to develop an improved classification, as well as to verify the practical significance of the research results by assessing their impact on minimizing potential losses for organizations. The work is based on a theoretical and methodological approach, incorporating a range of theoretical and analytical methods, such as systems and comparative legal analysis, content analysis of regulatory documents and scientific publications and formal logic modeling, which allowed for the structuring of the research. A comparative analysis of existing scientific studies on this topic in domestic and foreign literature revealed their limited applicability to the risks posed by AI and Big Data. The scientific novelty of the research lies in the development of a theoretical and methodological approach to improving the classification of compliance risks, based on the identification and systematization of new specific types of threats generated by the fundamental properties of AI systems (autonomy, variability, scalability), including latent discrimination and algorithmic distortion of factual data; as well as on the ontological substantiation of the classification criteria for specific types of compliance risks, which has expanded the scope of digital compliance and increased the risk coverage ratio from 42% (in basic models) to 90% (in the author's model). To confirm the practical validity of the theoretical and methodological principles of the proposed classification, the author developed an instrumental and computational block aimed at verifying the predictive effectiveness of the proposed measures. This novelty aspect lies in the development of a methodology for quantitatively assessing the effectiveness of the proposed solutions, including the substantiation of a comparative effectiveness coefficient, allowing for mathematical verification of the superiority of the author's approach over traditional static models; and the derivation of a proprietary formula for determining prevented potential damage, ensuring the conversion of qualitative risk indicators into measurable quantitative indicators of the organization's economic benefit. The classification of compliance risks proposed in this article can serve as a theoretical foundation for the creation of predictive compliance monitoring models, proactive adaptation of the regulatory framework, development of predictive, preventive control systems, digital compliance risk profiles and the automation of entity liability assessment.