The ratio of potential evapotranspiration (E0) over precipitation (P), known as the aridity index (AI), has been commonly used to stratify global aridity zones and widely adopted to assess changes in aridity globally. Anthropogenic climate change, in particular atmospheric warming, is projected to increase AI, which has in most cases been interpreted as increasing terrestrial aridity. In this study we demonstrate, for both past and future conditions, that such an interpretation requires reconsideration. Using catchment observations over the past 30 years and climate model projections for the 21st century, we show that increased AI does not ubiquitously lead to a decreased water availability over land, using surface runoff (Q) as an indicator. This is primarily caused by a higher Q sensitivity to changes in P (SP) and a lower Q sensitivity to changes in E0 (SE0), with the ratio of SP over SE0 being higher than the relative changes of E0 compared to P (i.e., |P × SP| > |E0 × SE0|). Assessment of Coupled Model Intercomparison Project Phase 5 model outputs indicates that both Q and AI change-induced Q changes are increasing over the majority of the globe for the 21st century despite increasing AI (a drying atmosphere). Our findings demonstrate a disconnection between the atmospheric drying trends and surface runoff trends and call for caution when interpreting retrospective and future changes in terrestrial aridity based on AI and related measures.