Predicting the sensory consequences of one’s actions is critical to perception and action in dynamically changing environments. For several established sensorimotor behaviours, such as locomotion and vocalizations, neural correlates of this prediction have been reported throughout sensory cortex. Oftentimes, however, the sensory consequences of an action are abstract, context-dependent, and newly learned, such as when one learns to play an instrument. How does cortex represent such complex sensorimotor rules? To investigate this question, we developed a task in which mice manipulated the frequency of an auditory cursor by licking either of two lick ports, in search for a rewarded target frequency. Trained mice monitored the ongoing stream of sounds to adaptively guide their lick behaviour. Using two-photon imaging, we then asked whether auditory cortical responses to tones reflected a representation of this novel, abstract sensorimotor rule, beyond purely sensory or movement-related effects. If animals did learn to predict the acoustic consequences of their licks, we reasoned that subtle violations of these predictions might modulate sound-evoked neural activity. Indeed, we find that frequency-tuned L2/3 excitatory neurons are sensitive to such violations, both at the level of the population of imaged neurons and at the level of individual neurons, even when matching stimulus history and licking. These effects were more pronounced in higher-order cortical subfield A2 than in the primary subfields A1 and AAF. Linear encoding models confirmed that neural responses to rule violations are better explained by sensorimotor prediction errors than by sensory or reward prediction errors. Together, these findings suggest that the relationship between actions and sensory feedback can shape responses along the sensory cortical hierarchy, even when this relationship is abstract and newly learned. Moreover, the continuous nature of the abstract foraging task presented here opens up new avenues for behavioural paradigms in animal models of cognition.