Subseasonal-to-seasonal (S2S) prediction is often described as a “predictability desert,” where the influence of initial conditions weakens while large-scale boundary forcings are partially developed. Yet improving S2S forecasts of extreme weather is essential for flood risk mitigation, water resource management, and societal resilience. With the rapid advancement of machine learning (ML) and artificial intelligence (AI), new strategies are emerging to enhance predictability by capturing complex, nonlinear processes, including teleconnections, subseasonal variability (e.g., MJO, QBO), and multiscale interactions. This session invites contributions exploring AI/ML for S2S prediction of natural hazards, including but not limited to AI-based forecasting of extremes (e.g., atmospheric rivers, extreme precipitation, tropical cyclones, heatwaves); hybrid dynamical–AI approaches; multi-model ensemble methods; AI data-driven post-processing and bias correction; AI agents for S2S prediction; evaluation of forecast skill across weeks 2–6 and seasonal scales; and regional case studies linking S2S forecasts to decision-making and risk management.