Reliable and high throughput communication in Massive Multiple-Input Multiple-Output (MIMO) systems strongly depends on accurate channel estimation at the Base Station (BS). However, the channel estimation process in massive MIMO systems is vulnerable to pilot contamination attacks, which not only degrade the efficiency of channel estimation, but also increase the probability of information leakage. In this paper, we propose a defence mechanism against pilot contamination attacks using a deep-learning model, namely Generative Adversarial Networks (GAN), to detect invalid uplink connections at the BS. Training of the models is performed via legitimate data, which consists of received signals from valid users and real channel matrices. The simulation results show that the proposed method is able to detect the pilot contamination attack with 98% accuracy in the best scenario.