Cloud computing has a significant impact on information technology solutions for both organizations and researchers. Different users share critical data over the cloud where failures are normal rather than exceptional. Therefore, data fragmentation and data replication algorithms are useful to enhance data security. Three important questions need to be answered carefully: (1) Which files should be replicated; (2) how many appropriate new replicas should be placed; (3) where the new replicas should be stored. In this paper, we propose a CSO-based approach for secure data replication (SDR) that determines suitable data center for new replica by designing a smart fuzzy inference system with four inputs as centrality, energy, storage usage, and load. In addition, a high-quality knowledge base is designed to describe the fuzzy system of CSO algorithm. To obtain a higher level of security, we partition each popular file into several fragments with different sizes based on the ability of data centers. Then, these fragments are stored based on the T-coloring concept to prevent an attacker from determining the locations of the fragments. Consequently, SDR protects the data file without any encryption technique since each data center has a single fragment of a particular file and no meaningful data are achieved in a successful attack. We evaluate the proposed algorithm with CloudSim toolkit, and the experiments show that SDR strategy can reduce the total energy consumption and response time by 31% and 28% (on average) compared to other related algorithms, respectively. In terms of storage usage, effective network usage, hit ratio, mean latency, load variance, number of replications, efficiency, and bandwidth consumption, the obtained results indicate that our strategy outperforms previous replication methods by a significant margin. The main reason is that SDR successfully balances the trade-offs among objectives by the fuzzy system.