Shear coupling implies that all grain boundary (GB) migration necessarily creates mechanical stresses/strains and is a key component to the evolution of all polycrystalline microstructures. We present MD simulation data and theoretical analyses that demonstrate the GB shear coupling is not an intrinsic GB property, but rather strongly depends on the type and magnitude of the driving force for migration and temperature. We resolve this apparent paradox by proposing a microscopic theory for GB migration that is based upon a statistical ensemble of line defects (disconnections) that are constrained to lie in the GB. Comparison with the MD results for several GBs provides quantitative validation of the theory as a function of stress, chemical potential jump and temperature.