Recent studies based on the ensemble mean of Coupled Model Intercomparison Project 6 (CMIP6) General Circulation Models (GCMs, CMIP6-GCMs hereafter) reported a decline in drought frequency over South Asia in the projected future climate. Here using the simulations from 16 CMIP6-GCMs, we examine the potential causes of declining droughts in South Asia. We show that the projections based on the multimodel ensemble mean CMIP6-GCMs are not reliable over South Asia. The multimodel ensemble mean is influenced mainly by the low-skill GCMs, which show high bias in simulating the monsoon (June–September) season precipitation during the observed period (1951–2014). The low-skill GCMs show a higher (20–30%) increase in the convective precipitation with a rise in the global mean temperature under the warming (1.5°C, 2.0°C and 2.5°C worlds) climate. The GCMs with less bias (BEST-GCMs) in the monsoon season precipitation and better seasonal cycle representation show lower sensitivity of convective precipitation to rise in global mean temperature. BEST-GCMs exhibit significantly different projections in comparison to the multimodel ensemble mean from all 16 GCMs (ALL-GCMs). In contrast to ALL-GCMs, BEST-GCMs project an increase in the frequency of droughts in South Asia under the future climate. Therefore, the projected risk of droughts over South Asia under the 1.5°C, 2.0°C and 2.5°C warming levels is higher than previously reported based on the ensemble mean of CMIP6-GCMs. A projected increase in the drought frequency in South Asia will have considerable implications for agricultural production and water availability.