The intensification of rainfall that is expected under global warming has alarming potential to increase soil erosion. For empirically based soil erosion modelling studies, quantifying this requires understanding of climate change impacts on rainfall erosivity. In most cases, current research is limited because the methods used for simulating erosivity do not properly account for changes in rainfall intensity and occurrence. Furthermore, nearly all findings are based on a small number of models developed under the fifth phase of the Coupled Model Intercomparison Project (CMIP5) which is being superseded by the more recent CMIP6. In this study, we perform the first assessment of high-resolution rainfall erosivity using a large ensemble of dynamically downscaled CMIP5 and CMIP6 rainfall for the state of Queensland, Australia. Changes in erosivity over the 21st century are considered under an intermediate, high, and very-high emissions scenario. We found that erosivity is projected to increase across most of the state and more so under higher emissions. The sharpest increases are predicted for the desert and grassland climate regions, while decreases are projected near the coast in the equatorial and tropical climate regions. Predicted changes in end-century erosivity and ensemble spread are smaller for CMIP6 than CMIP5, although changes in ensemble mean erosivity were not significantly different between CMIP phases because of the relatively high ensemble uncertainty. Changes in mean annual rainfall, followed by wet-day rain intensity and the 99th percentile intensity were the main drivers of change, although the importance of different rainfall attributes varied regionally. This study highlights the value of combining ensembles of daily regional climate projections with suitable erosivity models for understanding variability in erosivity impacts and mechanisms across different regions.