The most recent Intergovernmental Panel on Climate Change assessment report concludes that the Atlantic Meridional Overturning Circulation (AMOC) could weaken substantially but is very unlikely to collapse in the 21st century. However, the assessment largely neglected Greenland Ice Sheet (GrIS) mass loss, lacked a comprehensive uncertainty analysis, and was limited to the 21st century. Here in a community effort, improved estimates of GrIS mass loss are included in multicentennial projections using eight state‐of‐the‐science climate models, and an AMOC emulator is used to provide a probabilistic uncertainty assessment. We find that GrIS melting affects AMOC projections, even though it is of secondary importance. By years 2090–2100, the AMOC weakens by 18% [−3%, −34%; 90% probability] in an intermediate greenhouse‐gas mitigation scenario and by 37% [−15%, −65%] under continued high emissions. Afterward, it stabilizes in the former but continues to decline in the latter to −74% [+4%, −100%] by 2290–2300, with a 44% likelihood of an AMOC collapse. This result suggests that an AMOC collapse can be avoided by CO2 mitigation.

The air–sea exchange of heat and carbon in the Southern Ocean (SO) plays an important role in mediating the climate state. The dominant role the SO plays in storing anthropogenic heat and carbon is a direct consequence of the unique and complex ocean circulation that exists there. Previous generations of climate models have struggled to accurately represent key SO properties and processes that influence the large-scale ocean circulation. This has resulted in low confidence ascribed to twenty-first-century projections of the state of the SO from previous generations of models. This analysis provides a detailed assessment of the ability of models contributed to the sixth phase of the Coupled Model Intercomparison Project (CMIP6) to represent important observationally based SO properties. Additionally, a comprehensive overview of CMIP6 performance relative to CMIP3 and CMIP5 is presented. CMIP6 models show improved performance in the surface wind stress forcing, simulating stronger and less equatorward-biased wind fields, translating into an improved representation of the Ekman upwelling over the Drake Passage latitudes. An increased number of models simulate an Antarctic Circumpolar Current (ACC) transport within observational uncertainty relative to previous generations; however, several models exhibit extremely weak transports. Generally, the upper SO remains biased warm and fresh relative to observations, and Antarctic sea ice extent remains poorly represented. While generational improvement is found in many metrics, persistent systematic biases are highlighted that should be a priority during model development. These biases need to be considered when interpreting projected trends or biogeochemical properties in this region.

We use two coupled climate models, GFDL‐CM4 and GFDL‐ESM4, to investigate the physical response of the Southern Ocean to changes in surface wind stress, Antarctic meltwater, and the combined forcing of the two in a pre‐industrial control simulation. The meltwater cools the ocean surface in all regions except the Weddell Sea, where the wind stress warms the near‐surface layer. The limited sensitivity of the Weddell Sea surface layer to the meltwater is due to the spatial distribution of the meltwater fluxes, regional bathymetry, and large‐scale circulation patterns. The meltwater forcing dominates the Antarctic shelf response and the models yield strikingly different responses along West Antarctica. The disagreement is attributable to the mean‐state representation and meltwater‐driven acceleration of the Antarctic Slope Current (ASC). In CM4, the meltwater is efficiently trapped on the shelf by a well resolved, strong, and accelerating ASC which isolates the West Antarctic shelf from warm offshore waters, leading to strong subsurface cooling. In ESM4, a weaker and diffuse ASC allows more meltwater to escape to the open ocean, the West Antarctic shelf does not become isolated, and instead strong subsurface warming occurs. The CM4 results suggest a possible negative feedback mechanism that acts to limit future melting, while the ESM4 results suggest a possible positive feedback mechanism that acts to accelerate melt. Our results demonstrate the strong influence the ASC has on governing changes along the shelf, highlighting the importance of coupling interactive ice sheet models to ocean models that can resolve these dynamical processes.