Detecting communities plays a crucial role in social network analysis, offering valuable insights into network structures and relationships. However, there is a pressing concern regarding the potential exposure of sensitive or personal information about individuals when analyzing these communities. To tackle these privacy challenges, we present a novel community deception approach which could hide community information for individuals by perturbing network structures. In particular, we first conduct sampling to obtain subnetworks from the original network. Subnetwork sampling allows us to work with large networks. Then, we propose pairwise constraints into subgraph autoencoders to perturb the connections in the original network. Finally, we employ a genetic algorithm to seamlessly combine the perturbed subnetworks and construct a complete ruined network, which could ensure that sensitive information remains concealed within the network. Extensive experiments on several real-world datasets demonstrate the effectiveness of our approach.