Abstract: Traditional methods for community detection attacks primarily focus on unsigned networks, leading to challenges such as the neglect of signed attributes, low algorithmic efficiency, and suboptimal quality when applied to signed networks. This study proposes a novel algorithm for community detection attack in signed networks. The algorithm introduces an individual encoding method that incorporates signed attributes and attack operations. During the evolutionary process, the fitness value is computed without relying on detection algorithms, thereby simplifying the attack process and improving algorithmic efficiency. Additionally, a local strategy based on the modularity of attacking signs is designed to enhance the algorithm's attack performance. The proposed optimization strategy was adapted to various attack frameworks, and the universality, robustness, and transferability of the attack algorithm were validated on both model networks and empirical real-world networks.