Rice production failure is a major threat to food security and supply chain resilience across India. In this paper, we examine future rice production failure risks across India by integrating down-scaled climate projections with machine learning models that capture complex crop-climate interactions. First, we identify key drivers of historical crop failures and demonstrate the critical role of monthly weather variability. We then use our historically-trained ML models to project future production failure risk across India, linking spatial changes in failure risk with projected future weather distributions and exploratory scenarios of changes in irrigation access. Our findings indicate that district-level risks of rice production failures in India are projected to increase (26% average) under future climate change across all shared socioeconomic pathways in the near (2025 - 2054) and far (2055 - 2084) future. Our analysis demonstrates that expanding irrigation access could play a vital role in mitigating these risks, with substantial risk reductions observed in high-vulnerability regions. These insights provide actionable information for policymakers aiming to enhance agricultural resilience, identifying priority areas where adaptive measures, particularly irrigation improvements, can most effectively reduce vulnerability to climate-driven production risks.