Motor units (MUs), which compose spinal motor neurons and their innervated muscle fibers, are the fundamental unit of motor control. For nearly one hundred years, researchers have studied how MUs are recruited during movement, and the textbook understanding is that a simplified control strategy is used. A single ‘common drive’ is sent to a muscle’s motor neuron pool, which controls the amount of force that muscle will produce. This understanding has drawn primarily from experiments involving subjects generating static or gradually changing forces. However, recent results, using experiments with a much wider range of force profiles, have shown otherwise. Multiple degrees of freedom were needed to explain MU activity across a wide range of movement conditions. A large question remains, however – what underlies this flexibility of MU control? Is it driven by the cortex? To answer that question, we analyzed simultaneous recordings of neurons in primary motor cortex (M1) and MUs in a monkey making a wide range of force profiles. We developed a model framework in which MUs were predicted from M1 neurons through a low-dimensional bottleneck, to test how many degrees of freedom within M1 were beneficial for predicting MU activity. If M1 is simply sending a ‘common drive’ to control all the MUs within the motor neuron pool, then a single degree of freedom within M1 (a 1-d bottleneck) would explain the MU activity equally well as using more degrees of freedom. However, models with a one-dimensional bottleneck were not able to accurately predict all MUs, particularly those selectively active during high-frequency oscillatory conditions. Rather, multiple dimensions within the bottleneck were beneficial, and model perturbation analyses demonstrated that subsequent latent dimensions (beyond the first) were selectively used for high-frequency movements. This strongly suggests cortical flexibility over MU control.