MIT researchers have created a machine-learning algorithm that generates plausible worst-case scenarios for natural disasters and market crashes without requiring historical extreme data. This new modeling method helps planners better prepare for unprecedented future risks by projecting the potential intensity and impact of rare extreme events.