In order to enable federated learning to meet higher security and efficiency requirements, a zero trust model using double encryption and batch encryption was proposed. Firstly, double encryption was used to prevent multi-party threats from the server and other participants. By selecting different encryption methods and setting the encryption order, the federated learning model can be guaranteed to operate normally in a more secure environment. Secondly, the batch processing module was introduced on the basis of double encryption. Through splitting and splicing operations based on the number of key bits, the efficiency of encryption was improved to ensure the normal operation of the federated learning model in a more efficient manner. Theoretical analysis and experimental results show that the proposed federated learning model of zero trust mechanism can prevent inference attacks from multiple parties, and maintain the overhead similar to that of single-layer homomorphic encryption. It can be seen that the application of zero trust mechanism in federated learning has a certain degree of feasibility, and can meet the requirements of high security and high efficiency at the same time.