Seasonal predictions of Arctic sea ice conditions often rely on statistical relationships of a set of predictors that have shown skill for historical conditions. However, with rapid changes occurring in the Arctic climate, it is unclear whether these statistical relationships will remain valid. Here, preindustrial control, present-day control, and 20th–21st century climate model integrations are used to assess predictors for end-of-summer ice extent under various and changing climate conditions. Of importance for future forecasting systems, we find that the variance of September extent anomalies explained by winter-spring sea ice predictors, such as the area of Arctic basin thin ice cover, increases during the transition to a seasonally ice-covered Arctic. In contrast, summer atmospheric circulation variability plays a decreasingly important role in explaining the end-of-summer ice cover anomalies. These changes are primarily related to climate-dependent changes in the location of summer sea ice anomalies.