This study evaluates the seismic safety of intermediate reinforced concrete moment frames subjected to single and sequential earthquakes, with emphasis on the application of both real and synthetic ground motions, as well as the development of a predictive model based on artificial neural networks (ANN). In the first step, 64 buildings were designed and modeled using ETABS. The trained ANN model predicted preliminary design parameters with reasonable accuracy, and comparative results using linear regression also demonstrated acceptable performance. In the second step, three frames of 3, 6, and 11 stories were subjected to 10 ground motion records and analyzed using incremental dynamic analysis (IDA) and nonlinear static pushover analysis. Synthetic ground motions compatible with the design spectrum were generated and compared with real records. The results indicated that synthetic records, due to their spectral compatibility, yield lower response dispersion and more stable mean behavior, whereas real records, owing to their specific frequency content and near-fault pulses, reduce the median collapse capacity by 15–20%. Seismic performance indices including collapse capacity, fragility curves, and behavior factor were calculated. The collapse capacity of the frames decreased due to damage accumulation from the first shock, and the mean behavior factor in sequential scenarios showed a 12% reduction. An optimal ANN model with two hidden layers comprising 9 and 10 neurons was designed and trained, capable of predicting the behavior factor under the most critical sequential earthquake scenarios with an average error of less than 6%. Overall, this study highlights the pivotal role of neural networks in seismic safety assessment and underscores the necessity of incorporating sequential earthquake effects in seismically active regions such as Iran.