AI-Driven Tax Fraud Detection and Tax Compliance Analytics: A Review of Methods, Challenges, and Responsible Deployment

Abstract

Tax authorities increasingly use artificial intelligence (AI) and machine learning (ML) to detect fraudulent refund claims, identity-related fraud, and anomalous financial reporting. This narrative review examines recent research on these methods and the problems that come with deploying them in tax administration. It covers supervised learning, unsupervised anomaly detection, graph neural networks, and privacy-preserving transformer models. It then discusses three groups of challenges: data quality and fragmented fiscal records, the black-box nature of complex models, and the ethical and legal duties of public agencies, including privacy, fairness, and accountability. The reviewed studies show that ML can improve audit selection and surface new fraud patterns, with one study of Italian audits reporting gains of up to 39% in detected evasion. They also show risks: audit selection that ignores race can still produce large racial disparities in the United States, and the Dutch childcare benefits case shows what happens when a risk model is opaque and unchecked. The review concludes that responsible deployment needs explainable models, fairness testing throughout development, privacy protections, and human oversight of final decisions. The article follows no formal search protocol, and future work should test these conclusions with a systematic review.