Slope stability analysis is a fundamental challenge in geotechnical engineering. This study presents an efficient probabilistic modeling approach for slope stability using stochastic artificial neural networks. Key geotechnical parameters including unit weight, cohesion, internal friction angle, Poisson’s ratio, elastic modulus, and slope angle were modeled as random variables. Training data were generated via finite element analyses with 4000 simulations to compute the safety factor.Two neural network models were developed: the first predicting the factor of safety, and the second predicting the slope angle. Various hidden layer architectures were evaluated, and the optimal structures were selected based on minimum prediction error. A linear activation function was employed to facilitate integration within a probabilistic framework, enabling application of the model in the first-order second-moment reliability method. Results revealed that the optimal architecture for the first model includes five hidden layers with 5, 1, 13, 8, and 14 neurons, while the second model’s optimal structure consists of five layers with 5, 5, 7, 7, and 8 neurons. It was also demonstrated that increasing network complexity does not necessarily enhance model performance. The proposed approach was applied to two case studies: the Yasuj–Kakan road slope in Iran and the Wozeka–Gidole road slope in Ethiopia. For the first case, the reliability index and failure probability were 2.42 and 0.0078, respectively; for the second, these values were 4.34 and 6.8124^10-6 .According to the performance level criteria, these results correspond to performance levels ranging from “above average” to “good” for both sites.