The interplay of random phenomena and continuous dynamics deserves increased attention, especially in the context of wireless sensing and control applications. Safety verification for such systems thus needs to consider probabilistic variants of systems with hybrid dynamics. In safety verification of classical hybrid systems, we are interested in whether a certain set of unsafe system states can be reached from a set of initial states. In the probabilistic setting, we may ask instead whether the probability of reaching unsafe states is below some given threshold. In this paper, we consider probabilistic hybrid systems and develop a general abstraction technique for verifying probabilistic safety problems. This gives rise to the first mechanisable technique that can, in practice, formally verify safety properties of non-trivial continuous-time stochastic hybrid systems. Moreover, being based on abstractions computed by tools for the analysis of non-probabilistic hybrid systems, improvements in effectivity of such tools directly carry over to improvements in effectivity of the technique we describe. We demonstrate the applicability of our approach on a number of case studies, tackled using a prototypical implementation.