This letter researches the resource allocation issue in an intelligent reflecting surface (IRS)-assisted cooperative cognitive covert communication system, where IRS relay efficiently secures the covert transmissions of secondary users from potential eavesdroppers. Taking practical considerations into account, it is assumed that while the IRS only acquires channel distribution information (CDI), it remains unaware of eavesdropper’s specific detection threshold. In this scenario, we introduce a transferred generative adversarial network based resource allocation algorithm (TGAN-RA), which comprises of a source domain generator, a target domain generator, and a discriminator. The proposed TGAN-RA has extracted and transferred resource allocation feature of secondary users not transmitting covert message, and then transformed the whole covert communication process into an interactive game between the legitimate users and the eavesdropping. Numerical results indicate that even under non-ideal conditions with only known CDI and unknown eavesdropper’s detection thresholds, the proposed TGAN-RA algorithm can effectively attain nearly optimal resource allocation for covert communication while ensuring rapid convergence.