As the global economy develops rapidly, financial fraud has occurred frequently, which not only seriously damages the interests of investors, but also disrupts the normal order of the capital market. Traditional financial fraud risk identification methods often fail to realize ideal results when faced with problems such as data imbalance and complex features. To this end, this paper conducts research on financial fraud risk identification based on generative adversarial network (GAN). First, a comprehensive data set containing financial and non-financial indicators is constructed, and then, the learning ability of the generative adversarial network for fraud data features is enhanced by improving the structure of the generative adversarial network. At the same time, cross-validation and parameter optimization strategies are utilized to enhance the model’s generalization ability. Experimental results show that the accuracy rate reaches 0.92, which is 8.2%, 5.7%, and 4.5% higher than 0.85 of the LR algorithm, 0.87 of the SVM algorithm, and 0.88 of the RF algorithm, respectively. It effectively solves the recognition problem caused by data imbalance and provides a new technical path for financial fraud risk identification.