Previous analyses of the possibility of global breadbasket failures have extrapolated risks based on historical relationships between climate and yields. However, climate change is causing unprecedented events globally, which could exceed critical thresholds and reduce yields, even if there is no historical precedent. This means that we are likely underestimating climate risks to our food system. In the case of wheat, parts of the USA and China show little historical relationship between yields and temperature, but extreme temperatures are now possible that exceed critical physiological thresholds in wheat plants. Novel UNprecedented Simulated Extreme ENsemble (UNSEEN) approaches use large ensembles to generate plausible unprecedented events, which can inform our assessment of the risk to crops. We use the UNSEEN approach with a large ensemble of archived seasonal forecasts to generate thousands of plausible events over the last 40 years, and compare the results with historically observed extreme temperature and precipitation. In Kansas, USA, extreme temperatures that would have happened approximately 1-in-100-years in 1981 now have a return period of 1-in-10 years, while in China, the current return period is on the order of 1-in-20 years. Record-breaking years exceeding critical thresholds for enzymes in the wheat plant are now more likely than in the past, and these record-breaking hot years are associated with extremely dry conditions in both locations. However, recent temperature extremes, especially in Kansas, have not kept pace with the change in risk, and we find the region has “gotten lucky” in recent years with a lower trend in extremes in the observations than in the UNSEEN results. We find that there is a high potential for surprise in these regions if people base risk analyses solely on historical datasets, and we characterize plausible extremes from the UNSEEN ensemble that can be used to inform adaptation planning in these regions.