Cognitive control guides behavior by controlling what, where, and how information is represented in the brain. Several mechanisms have been proposed to support cognitive control, such as gain modulation or synchronizing neural activity. In this talk, I will present evidence for a new mechanism of cognitive control: by dynamically transforming neural representations into and out of different ‘subspaces’, cognitive control can route information through the brain. Large-scale recordings in mice and monkeys found neural activity within each brain region existed in a high-dimensional space. By recording from multiple regions simultaneously in mice, we found within each region’s high-dimensional space there was a subset of dimensions that was shared with other brain regions. When the neural response within a region was aligned to that subspace, neural activity propagated to the associated region(s). This may provide a mechanism for cognitive control – by aligning neural representations with different subspaces, information can be routed in different ways. Consistent with this, neural recordings in monkeys found different tasks transformed the neural representation of task-relevant information into two different subspaces in order to drive behavior. Altogether, our results suggest the dynamic transformation of information may be an important mechanism for controlling what and how neural representations are acted upon.