Ancient masonry structures represent a fundamental part of cultural heritage, embodying historical, architectural, and social significance. However, their vulnerability to various damage mechanisms, such as long-term material degradation and seismic cracking, poses significant preservation challenges, especially within the dynamic context of urban environments. To ensure the longevity of these structures and minimize the risk of critical events and structural failures, the development of advanced monitoring strategies is essential. Following the growing interest in Artificial Intelligence, this paper focuses on the application of an autoencoder-based strategy for damage assessment of the San Pietro bell tower, located in Perugia, Italy. Unsupervised anomaly detection is performed using real acceleration measurements recorded by the installed Structural Health Monitoring system, which captured the structural response to the seismic events that occurred in 2016. The goal is to demonstrate the potential of the implemented network in identifying data anomalies and promptly providing an alarm when observing abnormal variations of the monitored features. Therefore, such machine-learning aided technology can be considered a valid solution to enable continuous assessment, with limited computational efforts, of large-scale masonry structures and a proactive approach to structural health management.