Passive sensing based on UHF RFID has increasingly proven its efficacy in various applications. However, existing security methods often face challenges in implementation on passive tags or fail to meet the high data read rates required by array sensing operations. To address these constraints, we introduce a novel security framework called PSC-tags, which is in line with the zero-trust security concept and well-suited for parallel deployment within RFID sensing array scenarios. PSC-tags leverages the phase sequence similarities inherent in tag arrays, leading to the development of an optimal tag group selection strategy. Concurrently, we customize a convolutional neural network incorporating an attention mechanism for authentication. This method is applied to XRF55, which is a comprehensive dataset of human indoor activities, as well as a sub–dataset collected in real–world scenarios. Extensive experimental results demonstrate the effectiveness of PSC-tags, with an average accuracy of 98.5% and only 56 milliseconds of authentication time per sample required. Notably, PSC-tags is compatible with commercial off-the-shelf (COTS) devices and does not require any additional data acquisition. The method significantly fortifies the defenses against multiple attacks within RFID array sensing.