Guaranteeing the security of transactional systems is a crucial priority of\nall institutions that process transactions, in order to protect their\nbusinesses against cyberattacks and fraudulent attempts. Adversarial attacks\nare novel techniques that, other than being proven to be effective to fool\nimage classification models, can also be applied to tabular data. Adversarial\nattacks aim at producing adversarial examples, in other words, slightly\nmodified inputs that induce the Artificial Intelligence (AI) system to return\nincorrect outputs that are advantageous for the attacker. In this paper we\nillustrate a novel approach to modify and adapt state-of-the-art algorithms to\nimbalanced tabular data, in the context of fraud detection. Experimental\nresults show that the proposed modifications lead to a perfect attack success\nrate, obtaining adversarial examples that are also less perceptible when\nanalyzed by humans. Moreover, when applied to a real-world production system,\nthe proposed techniques shows the possibility of posing a serious threat to the\nrobustness of advanced AI-based fraud detection procedures.\n