<p>This paper investigates the probability of failure of shallow foundations under centered loading and in undrained conditions, designed with the Load and Resistance Factor Design (LRFD) method. The study uses Random Finite Element Limit Analysis (RFELA), accounting for the spatial variability of undrained shear strength. A recently developed RFELA framework is employed to efficiently evaluate bearing capacity. The method combines upper bound finite element limit analysis with random field theory. It generates representative trial collapse mechanisms and maps plastic dissipation across the spatial domain, allowing for the extrapolation of collapse loads to millions of random field realizations. This strategy allows sufficiently accurate estimation of failure probabilities at small additional computational cost. Several design scenarios are systematically analyzed, considering variability of the undrained shear strength, spatial correlation lengths, load ratios, and sets of partial factors. The study examines different approaches for determining characteristic soil properties and tested two alternative methods for estimating failure probabilities. The results show that proper local soil characterization for determining characteristic values of the undrained shear strength has a relevant impact on design reliability. Furthermore, the findings indicate that standard partial factors may lead to inconsistent probabilities of failure. A new set of partial factors is proposed, providing more consistent failure probabilities across different load ratios and soil variability.</p>