Abstract Recent decades have seen persistent sea surface temperature (SST) cooling in the Southeast Pacific (SEP) and Southern Ocean (SO), contrasting with broad ocean warming expected under anthropogenic forcing. Using an interpretable machine‐learning attribution framework applied to multi‐source observations and reanalyses, we focus on monthly SST variations beyond intrinsic oceanic persistence and relate them to physical predictors. Cooling in each region is associated with distinct combinations of physical predictors. In the SO, enhanced sea‐ice variability and lower‐tropospheric circulation anomalies account for a substantial fraction of the post‐2000 cooling. In the SEP, cooling is associated with a strengthened subtropical high, thermodynamic and cloud–radiative feedbacks. A moving‐window analysis indicates asymmetric coupling, with strengthening SO‐to‐SEP association in SST variability and a relative decline in local SEP feedbacks. These results show how high‐latitude–subtropical interactions can modulate regional SST trends under global warming through circulation, sea ice, and air–sea coupling.