This study proposes an optical multiimage authentication method for attenuating crosstalk noise in multiplexed images and addressing information security problems at different authentication levels based on hyper-chaotic amplitude masks and phase information multiplexing in the Gyrator transform domain. According to the proposed method, hyper-chaotic random amplitude masks are constructed first using a fractional hyper-chaotic Rabinovich system. Then, original images are encoded using a modified Gerchberg-Saxton algorithm for low- and high-level authentication. The hyper-chaotic masks are used as the amplitude constraint to iteratively obtain target images. The obtained N target images are then encoded into a composite image. Finally, the composite image is converted into two phase-only masks for transmission using the modified Gerchberg-Saxton algorithm. Results demonstrate that users can have their own authentication keys in the process of authentication at different security levels. In low-level authentication, the correctness of the authentication image can be determined by retrieving the nonlinear correlation peak between the authentication and original images. In high-level authentication, the authentication image can be obtained to be highly similar to the original image with the peak signal-to-noise ratio and correlation coefficient reaching 25.2817 dB and 0.9844, respectively. The proposed method is robust against occlusion and noise attacks and provides a new idea for optical authentication of multiple images at different levels.