<p>Phishing is a significant threat in social media, involving the deliberate dissemination of malicious URLs to steal private information and trick users. To combat this issue, various communities have developed detection methods to improve the identification of phishing attempts. This study introduces an innovative approach, instead of using traditional deep learning approaches to deal with phishing detection challenges, called Generative Round Network (GRN). It is a variant of the standard Generative Adversarial Network (GAN) framework by incorporating a Stochastic Rounding Function within the neural network components. The research utilizes a comprehensive dataset comprising both legitimate and phishing URLs to construct the classifier. The research finding shows that the proposed GRN model can produce a generator loss amount of 8.7319, indicating the cost associated with the generator’s ability to produce realistic data and a discriminator loss for generated data amount of 0.0212, while the discriminator loss for real data is 0.0714. The score difference of losses with GRN is 0.0502 means that the discriminator is better at distinguishing real data from generated data.</p>