Over the last decade, machine learning security has become an essential field of study. Researchers developed attacks aimed at compromising data-driven systems’ integrity, availability, and privacy and used them to assess the security of machine learning systems. However, most of these attacks have been studied in the domain of image recognition, and their applicability in security applications remains unclear. Despite its economic relevance, the research community has largely ignored credit card fraud detection, also known as fraud detection, and only a few solutions have been developed in this field. This work aims to provide a guide to understanding adversarial attacks in the context of fraud detection. We begin with an overview of the challenges posed by credit card fraud detection and adversarial machine learning, introducing the concept of threat modeling as a framework for constructing a taxonomy of adversarial strategies and declining it to the fraud detection context. Next, we review the main types of attacks in the literature, explore how these techniques can be applied in the context of fraud detection, and examine the solutions proposed to bridge the gap. Finally, we discuss promising research directions, drawing inspiration from the literature on other threat detection systems to discuss possible approaches to better understanding the domain.