Convolutional neural networks (CNNs) are good at putting known types of malware into groups, but they’re not so good at finding subtle differences and new threats, especially when data is scarce or there are imbalances between classes. To overcome these challenges, researchers have integrated generative adversarial networks (GANs) to augment training datasets and enhance the diversity of malware samples. In this study, we investigate the performance of hybrid GAN-CNN models compared to traditional CNN-based approaches in malware classification tasks. Through empirical evaluations across diverse malware datasets, we assess the accuracy, robustness, and adaptability of these models, with a focus on their ability to detect novel threats. The results of our study offer significant contributions to the practical effectiveness of hybrid GAN-CNN architectures in concrete cybersecurity scenarios. Quantitative evaluation measures, including accuracy, precision, recall, and F1-score, are employed to compare the effectiveness of the proposed method with the most advanced techniques currently available.