Learning in neuronal circuits requires coordinated modifications of synaptic strengths across hierarchical networks to refine computation and improve behavioral outcomes. This coordination requires solving the credit assignment problem. Yet, how neuronal circuits achieve credit assignment remains a central unsolved question in systems neuroscience [1]. Several modeling studies have suggested putative biological mechanisms for back-propagating error signals through multi-layer networks. These functionally motivated models assume distinct neuronal compartments representing local error signals that dictate the sign of synaptic plasticity [2-4]. However, this explicit error modulation is inconsistent with experimental findings [5,6] and phenomenological plasticity models [7] in which the sign is determined by a postsynaptic activity threshold. Here, we demonstrate that an inhibition-controlled Hebbian learning rule embedded in a plausible microcircuit model can resolve this discrepancy between normative and phenomenological models of plasticity. Inspired by experiments highlighting the control of inhibitory microcircuits over excitatory plasticity [8,9], we propose an inhibition-controlled plasticity (ICP) rule, in which inhibitory currents directly affect the plasticity threshold similar to previous work [10,11]. When embedded in a microcircuit with recurrent inhibition, ICP creates a stable fixed point for synaptic weights that naturally stabilizes Hebbian learning. At this fixed point, deviations from the expected inhibitory current dictate the sign of plasticity. In our model, receptive fields of excitatory neurons naturally de-correlate through lateral inhibition without the need for inhibitory plasticity. Finally, assuming top-down dis-inhibitory afferents carry credit information, ICP can be derived from first principles within an adaptive control theory framework and naturally performs error-modulated learning. Our empirical results show that such dis-inhibitory control performs comparably to backpropagation on several non-linearly separable benchmarks. In summary, we propose a Hebbian learning rule with an inhibitory contribution to the plasticity threshold that unifies the learning dynamics of phenomenological plasticity rules and normative models of gradient-based learning. Our findings make concrete predictions on inhibitory modulation of excitatory plasticity.