This paper systematically reviews Generative Adversarial Networks (GANs)-based cybersecurity defense technologies from 2021 to August 31, 2025. Using the PRISMA protocol, we searched 829 initial records from five major digital libraries and selected 185 peer-reviewed studies for analysis. We present a four-dimensional taxonomy encompassing defense functions, GAN architecture, cybersecurity domains, and adversarial threat models, highlighting the effectiveness of GAN-based defenses in network intrusion detection, malware analysis, and IoT security. Despite the advancements in WGAN-GP, CGAN, and hybrid GAN models, we highlight challenges that remain, including training instability, lack of standardized benchmarks, high computational costs, and limited explainability. Finally, we present a roadmap that emphasizes hybrid models, integrated evaluation, real-world applications, and LLM-based cyberattack defense.
