Abstract: In response to the problems of fast disaster causing process, short prediction period, and difficulty in early warning of flash floods in complex terrain mountainous areas, the research progress of artificial intelligence technology in meteorological forecasting, rainfall monitoring, flood forecasting, and knowledge reasoning has been systematically summarized according to the temporal logic of "early warning-short-term warning-dynamic warning-risk warning". The research has found that Artificial Intelligence(AI) is driving a transition in flash flood early warning from empirical threshold-based identification toward multisource data fusion, process-mechanism coupling, and intelligent risk inference, thereby improving rainfall forecasting, flood simulation, and decision-support capabilities. However, flash flood warning in complex mountainous regions still faces major challenges, including the scarcity of extreme rainfall samples, difficulties in cross-basin process transfer, and constraints on operational deployment. Future research should strengthen collaborative modeling of meteorological, hydrological, and flashflood disaster-chain processes, develop physically constrained transfer learning methods, and improve risk governance for AIbased warning models. These advances will enhance the accuracy, generalization capacity, reliability, and practical applicability of flash flood early-warning models.