Extreme climate events or tails of natural hazard distributions tend to be the most damaging in terms of societal impacts. While traditional physics-based approaches are suitable for gaining mechanistic understanding and for process-based studies, they may not be adequate for characterizing probabilistic risk from extreme events. Here, we demonstrate a ML/AI-based approach for estimating probabilistic risk from tropical cyclones (TCs) and the associated coastal flooding. First, we simulate nearly one million TCs for the current and future climates using the Risk Analysis Framework for Tropical Cyclones (RAFT), a hybrid model that combines physics with deep neural networks. Subsequently, we apply RAFT-simulated TC tracks to DeepSurge, an AI-based storm surge model that is trained on a large number of ADCIRC simulations in the North Atlantic. Our results suggest a significant increase in storm surge risk for the northern Gulf Coast, Florida and some areas near the mid-Atlantic. Also importantly, we show that ML/AI can be leveraged effectively to address the potential issue of ‘Grey Swan’ TCs and their impacts.