Drought-driven hydroclimatic stress is increasingly disrupting agricultural supply in arid and semi-arid environments. This study quantifies future supply vulnerability for three major cereal crops in Iran, namely wheat, rice, and barley, using a multivariate Long Short-Term Memory (LSTM) time-series framework. Annual data for 1961–2023 were compiled from international databases and include crop production, imports, harvested area, and precipitation as a climate proxy. The modelling workflow proceeds in two steps: (i) domestic production is projected for 2024–2028 using crop-specific multivariate LSTM models; and (ii) the forecasts are translated into implied import requirements. In this study, implied import requirement is a gap-based accounting indicator, calculated as the non-negative difference between the calibrated baseline domestic requirement and forecasted production, rather than a direct machine-learning forecast of import volumes. To stress-test vulnerability, three drought shock scenarios, including 10%, 30%, and 50% production reductions, are applied to the baseline production pathway, and scenario-specific import gaps are recomputed. The results show that drought shocks substantially increase external supply exposure across all crops. Under the severe 50% shock scenario, average implied import requirements during 2024–2028 increase by 172.6% for wheat, 137.1% for rice, and 65.4% for barley compared with the baseline pathway. In aggregate, the three-crop import gap rises by 135.7%, indicating a sharp increase in food-security vulnerability under severe drought stress. Methodologically, the study demonstrates how LSTM-based production forecasts can be converted into policy-relevant import-gap scenarios. From a policy perspective, the findings support integrated drought-risk management through water productivity improvement, and forward import planning.