The CNS may produce the same endpoint trajectory or torque profile with different muscle activation patterns. What differentiates these patterns is the presence of co-contraction, which does not contribute to joint torque generation but allows to modulate mechanical impedance. Whether co-contraction is controlled through the same synaptic input to motor neurons involved in generating joint torque is still unclear. We hypothesized that co-contraction is controlled through a specific synaptic input, independent from that underlying the control of torque. To test this hypothesis, we asked participants to concurrently generate multi-directional isometric forces at the hand and to modulate the co-contraction of arm muscles to displace and stabilize a virtual end-effector. The firings of motor units were identified through decomposition of High-Density EMGs collected from two antagonist muscles, Biceps Brachii and Triceps Brachii. We found significant peaks in the coherence between the neural drive to the two muscles, suggesting the existence of a common input modulating co-contraction across different exerted forces. Moreover, the within-muscle coherence computed after removing the component synchronized with the drive to the antagonist muscle or with the exerted force revealed two subsets of motor neurons that were selectively recruited to generate joint torque or modulate co-contraction. This study is the first to directly investigate the extent of shared versus independent control of antagonist muscles at the motor neuron level in a task involving concurrent force generation and modulation of co-contraction.