AI RISKS IN BANKING AND APPROACHES TO THEIR ASSESSMENT

Abstract

The article systematizes the risks that arise in the process of implementing artificial intelligence (AI) technologies in banking and substantiates methodological approaches to their assessment. The concept of “AI risks in banking” is clarified, a detailed classification of AI risks is proposed (model, data, cybersecurity, operational, ethical, reputational, and regulatory risks), and their place within the bank’s overall risk system is outlined. Particular attention is paid to the interaction between AI-specific risks and traditional financial risks, such as credit, market, and liquidity risks, as well as to the risk of non-compliance with regulatory and supervisory requirements.

Based on a synthesis of international standards, regulatory guidelines, and academic literature, an integrated methodology for assessing AI risks is developed, combining quantitative (probability–impact matrices, key risk indicators, scenario analysis, and stress testing) and qualitative (algorithmic audit, bias and fairness assessment, explainable AI tools, model validation, and comprehensive model risk management frameworks) tools. The article emphasizes the importance of ensuring transparency and accountability of AI systems, defining clear lines of responsibility between business units, IT, and risk management divisions, and implementing continuous monitoring of AI models throughout their life cycle.

The study shows that effective AI risk management in banking requires institutionalizing model risk management functions, strengthening data governance and cybersecurity policies, enhancing staff competencies, and adapting regulatory requirements and internal policies to the specific nature, scale, and complexity of AI applications in the banking sector.