Parkinson patients suffer from levodopa-induced dyskinesia, which occur adversely to chronic dopaminergic treatment. These abnormal involuntary movements can only partly be actively suppressed and affect quality of life. A lowered motor inhibition during hyperdopaminergic states, associated with structural and plasticity changes in the cortico-basal-ganglia motor network, is hypothesized to enable dyskinesia. Multiple oscillatory cortico-subthalamic patterns associated with dyskinesia have been described but their dependence on behavioral states such as movement presence remains unknown, which is crucial for its use in real-life application of adaptive neuromodulation. We studied invasive cortico-subthalamic oscillations in 22 patients with Parkinson’s disease during dyskinesia-evoking protocols. Clinical assessments differentiated between non-dyskinetic and dyskinetic periods, and kinematic monitoring detected movement presence, leading to four behavioral states containing rest, voluntary movements, movement suppression during dyskinesia, and dyskinetic movements. Data-driven methods reduced data dimensionalities and optimized frequency-specific signal-to-noise ratios in the neural recordings and allowed feature extraction of spectral magnitudes, variances, and inter-subthalamic and cortico-subthalamic coherences. Subthalamic theta-activity and attenuated beta-activity were elevated during both dyskinetic movement suppression and execution, while cortico-subthalamic gamma-activity was only increased during dyskinetic movement execution. The subthalamic oscillations predicted dyskinesia presence, but varying behavioral states containing fluctuating movement presence affected the predictive performance. Movement-aware classifications improved dyskinesia detection based on cortical and on gamma oscillations. Introducing a movement-aware classification which considered the current behavioral state improved the neural detection of therapeutic states. We propose movement execution during dyskinesia should be considered as a distinct behavioral and neural microstate within a dopamine-depending hyperdopaminergic macrostate. Integrating this state concept may inform future adaptive neuromodulation and enhance its naturalistic robustness during every-day life.