This paper focuses on the assessment of earthquake risks associated with extreme losses caused by seismic events. The accurate fitting of thick-tailed data, precise evaluation of extreme risks, and formulation of effective preventive measures are crucial in reducing the frequency and impact of such events. The study utilizes historical earthquake cases in mainland China from 1980 to 2022 as a representative sample. The POT model, based on the generalized Pareto distribution, is employed to fit the thick-tailed losses resulting from earthquakes. The model diagnostic results indicate a good fitting effect. By estimating the model parameters and analyzing historical data, the study calculates the Value at Risk (VaR) values, which represent the maximum potential direct economic losses and casualties at different confidence levels in the event of an earthquake. Moreover, the average percentage of deaths and injuries in historical cases is used to estimate the maximum number of fatalities and injuries resulting from an earthquake. Through a comprehensive analysis of the three types of losses surpassing the maximum loss at different confidence levels, a seismic risk zoning system is proposed, along with corresponding preventive measures. The aim is to provide the government with scientifically grounded options for earthquake risk prevention and management, thereby mitigating seismic risks and ensuring the safety of lives and properties.