Abstract: The assessment of flood risk resulting from embankment dam breaching is essential for developing effective emergency response strategies and improving disaster relief efforts. Embankment dams, widely adopted for their cost-efficiency, construction simplicity, and adaptability to diverse terrains, are particularly susceptible to breaching under extreme hydrological conditions, structural deficiencies, or inadequate maintenance. When a breach occurs, the sudden release of stored water can trigger a catastrophic flood wave, causing severe consequences such as substantial loss of life, infrastructure damage, environmental degradation, and prolonged socio-economic disruption in downstream areas. Despite the urgency of accurate risk evaluation, existing methodologies for flood discharge simulation and loss assessment often exhibit critical limitations. Traditional models tend to oversimplify the complex breaching processes and inadequately capture the nonlinear dynamics of flood propagation. Additionally, many approaches lack robust techniques for estimating both direct and indirect losses and offer limited tools for quantifying predictive uncertainty. To address these challenges, this study introduces a novel mechanism-data dual-driven framework for flood simulation and loss assessment tailored to embankment dam breach scenarios. This method integrates detailed physical modeling of the dam breach process with data-driven statistical analysis of flood propagation and its consequences. By combining these complementary perspectives, the approach enables more realistic and accurate flood simulations, offering a stronger foundation for risk assessment. A central innovation of the methodology is the incorporation of a Bayesian statistical model, which facilitates probabilistic analysis and enables rigorous quantification of uncertainties in estimating both life loss and economic damage. The assessment framework consists of three interrelated modules. The first module simulates the breach process using a combination of physical models, empirical equations, and field data to reconstruct the timing, size, and evolution of the breach. The second module calculates key disaster-inducing factors during flood propagation—such as flow velocity, flood depth, inundation extent, and arrival time in affected areas—which are critical for evaluating the intensity and spatial reach of the event. The third module estimates human and economic losses by integrating the hydrodynamic outputs from the second module with demographic, land use, and infrastructure data to assess exposure and vulnerability. The Bayesian model is then employed to quantify uncertainty and generate probabilistic loss estimates. To validate the effectiveness and reliability of the proposed method, it was applied to the 2018 Sheyuegou dam breach case in China. This real-world event provided a comprehensive dataset for model calibration and verification. Inversion analysis results showed that the relative errors for key parameters—such as peak flood discharge, final breach dimensions, and timing of peak flow—were all within ±15% of observed values. Moreover, the maximum deviation in the highest water level at representative cross-sections within the inundation zone was only −0.21 meters. Regarding casualty estimation, the predicted number of fatalities closely matched the reported figures, underscoring the method’s accuracy and credibility. In conclusion, these findings confirm that the proposed mechanism-data dual-driven approach is an effective and practical tool for simulating dam breach scenarios and assessing associated flood risks. Its capacity to accurately replicate real-world outcomes while incorporating uncertainty in loss estimates provides valuable support for emergency preparedness, response planning, and disaster risk reduction. This research advances flood risk assessment methodologies by offering a more comprehensive, precise, and scientifically robust framework for managing the impacts of embankment dam failures.