Drought sensitivity, defined as ecosystem susceptibility to meteorological drought, is a key indicator of terrestrial ecosystem response to water stress and thereby critically regulates carbon sink function. However, little is known about the spatiotemporal evolution of drought sensitivity and its effects on carbon uptake changes. Here, we integrated multi-source remote sensing gross primary productivity data and meteorological drought indices in the Yellow River Basin from 1982 to 2018. Using interpretable machine learning methods, we revealed post-drought shifts in sensitivity, identified key drivers of the spatial changes in drought sensitivity, and quantified its influence on ecosystem carbon uptake. Our results showed that drought sensitivity of 57.12 % of the basin increased significantly (P < 0.05), particularly in arid regions and grassland ecosystems. Climate change was the main driving factor of drought sensitivity change, contributing 75.55 %, with rising solar radiation, intensifying drought, and increasing temperature as the primary factors. Increased drought sensitivity prolonged ecosystem recovery, reducing annual carbon uptake by 0.11 +/- 0.05 Tg C during the 3-year recovery period, further weakening the ecosystem carbon sink capacity. Our findings highlight that increasing drought could further amplify the potential risk of carbon sinks in terrestrial ecosystems.