Cooperation is intrinsic to the human ability to work together toward common goals, and depends on sensing and reacting to dynamically changing relationships between coacting partners. Using functional magnetic resonance imaging (fMRI) and a paradigm in which an adaptive pacing signal simulates a virtual partner, we examined the neural substrates underlying dynamic joint action. A single parameter controlled the degree to which the virtual partner adapted its behavior in relation to participant taps, thus simulating varying degrees of cooperativity. Analyses of fMRI data using objective and subjective measures of synchronization quality found the relative balance of activity in two distinct neural networks to depend on the degree of the virtual partner's adaptivity. At lower degrees of adaptivity, when the virtual partner was easier to synchronize with, cortical midline structures were activated in conjunction with premotor areas, suggesting a link between the action and socio-affective components of cooperation. By contrast, right lateral prefrontal areas associated with central executive control processes were recruited during more cognitively challenging interactions while synchronizing with an overly adaptive virtual partner. Together, the reduced adaptive sensorimotor synchronization paradigm and pattern of results illuminate neural mechanisms that may underlie the socio-emotional consequences of different degrees of entrainment success.