In recent years, artificial intelligence has been deeply integrated into various business scenarios of libraries, bringing new data security risks that traditional protection systems are difficult to cope with. Focusing on the current situation from three aspects including institutional norms, technical deployment and operational management, this paper summarizes three major core risks in libraries: data fusion risks, algorithm black box risks and external dependency risks. Corresponding countermeasures are put forward, such as deploying privacy computing technology, implementing algorithm security auditing, and optimizing supply chain management. The research findings can provide practical references for libraries to establish a proactive defense-oriented data security system, and boost the high-quality development of smart libraries through the improvement of data security governance capability.