Overview of modern methods for detecting financial crimes using artificial intelligence agents

DOI: 10.31673/2412-9070.2025.050674

Abstract

The article presents a comprehensive analysis of modern methods for detecting financial crimes using artificial intelligence (AI) agents. It examines the classification of AI agents (reactive, deliberative, autonomous, and multi-agent), their operational features, and their role in financial monitoring systems. A comparative analysis of rule-based approaches, machine learning methods, hybrid models, blockchain architectures, and graph algorithms has been conducted. The study reveals that hybrid solutions and graph neural networks demonstrate the highest levels of precision and recall in detecting suspicious transactions, as confirmed by consolidated metrics from peer-reviewed sources.
Special attention is given to the use of deep neural networks, time series processing techniques, natural language processing (NLP), and Explainable AI (XAI) methods, which enhance the transparency and interpretability of AI-driven decisions—an essential requirement in heavily regulated financial domains. The advantages of the multi-agent approach are emphasized, including the ability for parallel analysis of complex fraud schemes, dynamic adaptability, and system scalability, which position it as a promising direction for the development of distributed financial intelligence systems. Alongside the identified benefits, the study also outlines key challenges hindering real-world implementation: limited interpretability of deep learning models, the requirement for large volumes of high-quality and balanced data, high computational costs, and legal and ethical constraints related to the automated processing of sensitive financial information, especially under regulations such as GDPR and AMLD.
In terms of future development, the article highlights the potential for integrating AI agents with blockchain networks to ensure transactional transparency and immutability, applying quantum algorithms for processing complex financial graphs, and adopting edge computing for real-time anomaly detection on decentralized devices. Furthermore, the research underlines the importance of an inter disciplinary approach that combines expertise in artificial intelligence, cybersecurity, economics, and legal compliance in building robust and effective financial crime prevention systems. Thus, the findings confirm that effective financial crime detection today requires complex technological solutions based on AI agents with an emphasis on transparency, scalability, and regulatory compliance. The conclusions presented in this study hold both theoretical and practical value for the development of modern financial transaction monitoring systems and the formulation of policies in the field of digital financial security.

Keywords: financial crimes; artificial intelligence; artificial intelligence agents; machine learning; blockchain; graph analysis.