Large Language Models for Financial Fraud Detection: A Systematic Review of Methodologies, Performance, and Future Directions

Abstract Financial fraud continues to pose significant threats to global economic stability, with reported losses reaching un- precedented levels in recent years. The traditional detection methods for fraud are valuable; however they suffer from changing fraud patterns, extreme class imbalances and increased complexity in the fraud schemes. This systematic review analyzes 33 peer-reviewed studies, primarily published between 2023 and 2025 with foundational studies from 2010-2013 included for tra- ditional method comparison, examining how Large Language Models (LLMs) are transforming financial fraud detection across multiple domains including credit card transactions, fintech applications, trading systems, and insurance claims. The literature can be grouped into six (6) questions based on the six (6) areas of research that address the detection performance, architec- tural innovation, computing requirements, domain coverage, implementation challenges, and future directions for the area of fraud detection. Our analysis reveals that hybrid systems combining LLMs with traditional machine learning consistently out- perform standalone approaches, while identifying significant research gaps in forex market fraud detection despite its massive daily trading volume. This review presents the most significant advances in technical innovation including; Data Serialization Techniques, Multi-Agent Frameworks, and Retrieval-Augmented Generation Systems, which have helped advance the space of LLM based Fraud Detection. Additionally, this review provides practical guidance on how Financial Institutions can implement LLM solutions for fraud detection, as well as Priority Areas for Future Research such as Real-Time Processing Optimization, Cross-Domain Generalization, and Automated Regulatory Compliance. Ultimately, this review is intended to serve as a starting point for Researchers and Practitioners who wish to gain an understanding of and continue advancing the use of LLMs for fraud detection. Keywords Large Language Models, Financial Fraud Detection, Machine Learning, Deep Learning, Systematic Review, Credit Card Fraud, Fintech Security, Transformer Networks.