The aim of this paper is to employ explainable artificial intelligence methods to prioritize cyber risks. To achieve this, we compare alternative machine learning models that predict the evolution of cyber attacks while accounting for time dependence. Subsequently, we apply explainable AI techniques to identify the most significant types of cyber attacks based on their occurrence. We illustrate our approach using data from the Hackmanac website, which categorizes worldwide cyber attacks by severity. The empirical findings reveal two distinct periods: before and after 2023. In both cases, eXtreme Gradient Boosting emerges as the best-performing model. Furthermore, the most relevant explanatory features remain relatively consistent across different explainable artificial intelligence methods but vary over time, reflecting the recent evolution of cyber threats.