Based on the upper bound limit analysis and random field theory, this paper investigates the reliability of excavated slopes in undrained conditions within the framework of Monte Carlo simulation. The soil undrained shear strength is simulated as random fields. The effects of degree of non-stationarity and coefficient of variation on the slope stability are highlighted in the parametric sensitivity analysis. The results indicate that the slope reliability index increases with the increase of degree of non-stationarity and reduces as the coefficient of variation increases. Meanwhile, the soil spatial variability affects significantly the failure mechanism of slopes. Under the combination of random parameters considered in this paper, the failure mode is mainly of deep failure. With the increase of degree of non-stationarity and coefficient of variation, the probability of occurrence for the deep failure mode gradually reduces and the probability of occurrence for the toe failure and deep toe failure modes increase accordingly.