Exploiting flexibility of low-cost but uncertain demand-side resources attracts great attention in the recent research of the distribution system with special respect to reliability assessment and improvement. While various kinds of uncertainties and risks exist, such as the stochastic power of renewable energy resources (RES), etc., the literature overlooked the exogenous and endogenous uncertainties along with the utilization of these resources denoted as virtual energy storage (VES). Instead, this manuscript proposes a novel approach for reliability assessment and improvement of the distribution system in special consideration of the endogenous uncertainties in VES operations, also denoted as decision-dependent uncertainties (DDUs). The modified and unified probabilistic model of VES is first presented to reveal the dynamic process and human behavior of VES. Additionally, we extend our recent work on probabilistic optimization under DDUs to the reliability assessment problem which addresses the optimal load curtailment with the support of VES. Moreover, reliability indices are newly introduced to (1) calculate the response losses of VES and the consequence of overlooking these uncertainties, and (2) explain the difference between theoretical and practical reliability. Case studies based on the ground truth data verify that the available capacity of VES will gradually reduce (maximum of 40 %–60 %) because of DDUs and it reveals the overlooked risks by using “greedy” strategies in the previous models, while the best practical reliability performance (increased 5 %) is witnessed via the proposed methods. We further investigate four key factors affecting the reliability performance, i.e., probability level of chance constraints, DDUs levels, dispatch time period, and locations of VES. Eventually, reasonable suggestions are provided for better flexibility utilization of VES, and the system's reliability is further improved by around 30 %.