Financial fraud detection is a critical and complex task that has become increasingly important with the rise of digital transactions. The challenge lies in identifying fraudulent activities, which constitute only a small fraction of the overall data, leading to a significant class imbalance. Traditional machine learning techniques often struggle to detect fraud accurately due to this imbalance, resulting in high false negative rates and missed detections. To address this issue, this paper explores the use of Generative Adversarial Networks (GANs) to generate synthetic fraudulent data, thereby balancing datasets and enhancing the accuracy of detection models. We employ a Voting Classifier that integrates three base learners: Gradient Boosting, Decision Tree, and Random Forest. The results demonstrate a notable improvement in fraud detection, with our proposed approach achieving a detection accuracy of 97%. By leveraging GANs, we show significant improvements in precision, paving the way for more robust fraud detection systems. Additionally, we discuss the ethical implications and future prospects of GAN-based techniques in financial security.