The death toll and monetary damages from landslides continue to rise despite advancements in predictive modeling. The predictive capability of these models is limited as landslide databases used in training and assessing the models often have crucial information missing, such as underlying failure types. Here, we present an approach for identifying failure types based on their movements, e.g., slides and flows by leveraging 3D landslide topology. We observe topological proxies reveal prevalent signatures of mass movement mechanics embedded in the landslide's morphology or shape, such as detecting coupled movement styles within complex landslides. We find identical failure types exhibit similar topological properties, and by using them as predictors, we can identify failure types in historic and event-specific landslide databases (including multi-temporal) from various geomorphological and climatic contexts such as Italy, the US Pacific Northwest region, Denmark, Turkey, and China with 80 to 94 % accuracy. To demonstrate the real-world application of the method, we implement it in two undocumented datasets from China and publicly release the datasets. These new insights can considerably improve the performance of landslide predictive models and impact assessments. Moreover, our work introduces a new paradigm for studying landslide shapes to understand underlying processes through the lens of landslide topology.