The role of internal variability in generating an apparent link between autumn Barents-Kara sea ice (BKS) and the winter North Atlantic Oscillation (NAO) has been intensely debated. In particular, the robustness and causality of the link has been questioned by showing that BKS-NAO correlations exhibit nonstationarity in both reanalysis and climate model simulations. We show that the lack of ice observations makes analysis of nonstationarity using reanalysis questionable in the period 1950-1970 and effectively impossible prior to 1950. Model simulations are used to corroborate an argument that nonstationarity is nevertheless expected due to changes in the ice edge variability due to global warming. Consequently, changes in BKS-NAO correlations over time may simply reflect that the ice edge has moved, rather than that there is no causal link. We discuss potential implications for analysis based on coupled climate models, which exhibit large ice edge biases.