Distributed minimax optimization plays a crucial role in modern machine learning. However, this framework is highly susceptible to adversarial attacks, especially model poisoning attacks, where malicious participants inject harmful updates to compromise the learning process. Existing defense mechanisms primarily focus on traditional distributed learning in minimization problems. To address this challenge, we propose Robust-LSGDA, a novel distributed minimax learning algorithm designed to defend against model poisoning attacks. Our algorithm achieves an asymptotically optimal convergence rate of \(\mathcal {O} \left( \frac{1}{\sqrt{KT} } \right) \) , where K is the number of clients and T is the global maximum number of iterations. Furthermore, we establish a robustness guarantee for Robust-LSGDA by analyzing its certified radius, providing theoretical assurance of its defense capabilities.