Many state-of-the-art climate models do not simulate the Atlantic Water (AW) layer in the Arctic Ocean realistically enough to address the question of future Arctic Atlantification and its associated feedback. Biases concerning the AW layer are commonly related to insufficient resolution and exaggerated mixing in the ocean component as well as unrealistic Atlantic-Arctic Ocean exchange. Based on sensitivity experiments with FESOM1.4, the ocean-sea ice component of the global climate model AWI-CM1, we show that even if all impediments for simulating AW realistically are addressed in the ocean model, new biases in the AW layer develop after coupling to an atmosphere model. By replacing the wind forcing over the Arctic with winds from a coupled simulation we show that a common bias in the atmospheric sea level pressure (SLP) gradient and its associated wind bias lead to differences in surface stress and Ekman transport. Fresh surface water gets redistributed leading to changes in steric height distribution. Those changes lead to a strengthening of the anticyclonic surface circulation in the Canadian Basin, so that the deep counterflow carrying warm AW gets reversed and a warm bias in the Canadian Basin develops. An underestimation of sea ice concentration can significantly amplify the induced ocean biases. The SLP and anticyclonic wind bias in the Nordic Seas weaken the cyclonic circulation leading to reduced AW transport into the Arctic Ocean through Fram Strait but increased AW transport through the Barents Sea Opening. These effects together lead to a cold bias in the Eurasian Basin.