Abstract To address the security vulnerabilities inherent in cognitive radio technology amidst extensive control information and data exchanges, this paper introduces the use of unmanned aerial vehicle equipped with intelligent reflecting surfaces (UAV-IRS) as relay nodes to facilitate covert communications for cognitive users. Given the complexities and dynamics of the electromagnetic environment, which involve multiple interconnected and constantly changing communication resources, as well as high-dimensional data, we propose the NeuroEvolution of Augmenting Topologies based Resource Optimisation Algorithm Method (NEAT-ROAM). This advanced algorithm continuously optimizes and adapts the network structure in response to the external environment, ensuring robust and secure communication in dynamic and challenging scenarios. The algorithm not only provides superior covert communication performance but also exhibits fast convergence, which can be applied to the problem of assisting covert communication under mobile conditions using UAV-IRS.