Dikes and levees play a crucial role in flood protection. The main causes of levee failures are of geotechnical nature. Geotechnical failure modes are also the main contributors to the probability of failure of flood defenses such as levees due to the large uncertainties in ground conditions. Hence, information on ground conditions and soil properties is crucial in safety assessments and retrofitting designs of levees. The experience in practice with designs and risk assessments is that even with a substantial amount of site investigation and laboratory testing the uncertainties in soil properties often remain large. One way to further reduce uncertainties is to include past performance information. The present papers shows how information of survived loads in the past can be incorporated in reliability assessments by means of Bayesian posterior analysis using the same (physics-based) performance models as the prior reliability assessments. The incorporation of survival information leads to reduced uncertainties and probabilities of failure. Furthermore, an approximate approach using fragility curves is proposed, mainly to make problems with computationally expensive performance functions tractable. After briefly recapping Bayesian reliability updating, we provide examples for slope stability of (flood protection) dikes. The examples illustrate that posterior analysis enables us to reduce the large prior uncertainties in soil strength parameters by "eliminating" implausible samples or regions of the joint PDF of the strength properties using the evidence of survived load conditions (i.e. water levels). The results suggest that the effect of updating the probability of failure with survival evidence can be considerable, especially when the uncertainty in the strength properties dominates the reliability. Dominant strength uncertainties are rather typical in geotechnical engineering, as opposed to, for example, structural engineering, implying that using the method is promising also for other geotechnical applications.