Artificial intelligence in financial risk management: global trends, XAI and regulatory approaches in Kazakhstan

K. Zh. Abisheva, И. В. Селезнёва, Sh. D. Kydyrbayeva, Mаrina V. Shtiller

2025

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.46914/1562-2959-2025-1-4-376-395
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

The integration of artificial intelligence (AI) and machine learning (ML) into financial risk management is accelerating globally and in Kazakhstan. These technologies enhance forecast accuracy and automate key processes, while simultaneously posing challenges for model transparency, decision ethics, and regulatory oversight. According to the National Bank of the Republic of Kazakhstan, about 31% of financial organizations are already using AI, while among second-tier banks the share of users reaches 60%. At the same time, only a small portion of organizations have integrated AI across all key business functions, indicating that most market participants are at an early stage of digital maturity [1]. This study presents a 2010–2025 systematic literature review conducted under PRISMA principles, combining international evidence with the Kazakhstani context of applying AI/ML to various categories of financial risk (credit, market, operational, fraud/AML). Bibliometric and thematic analyses indicate a sharp post-2015 increase in publications, diffusion of complex architectures (deep neural networks, ensemble methods), and rising attention to explainable AI (XAI). Contemporary ML algorithms deliver substantial improvements in the accuracy, speed, and reliability of risk forecasts relative to traditional approaches [2], while limitations in interpretability and production-level implementation persist. The operational robustness of solutions requires MLOps practices (version control, automated deployment of software and models into production environments, and model monitoring and validation). The scientific novelty of the article lies in the comprehensive systematization of AI methods by types of financial risks (credit, market, operational, fraud/AML) with consideration of XAI, as well as the development of a structured roadmap for AI implementation for banks and regulators. The practical significance lies in a set of specific recommendations for the development of data infrastructure, model risk management processes, XAI tools, and SupTech solutions designed for use by financial organizations and supervisory authorities in Kazakhstan.