Paediatric status epilepticus (SE) is a medical emergency associated with significant morbidity. Benzodiazepines (BZPs) are the current first-line treatment, but do not work in over a third of children presenting with SE. Animal studies have shown that SE can cause changes in synaptic inhibition signalling which can ultimately lead to BZPs becoming ineffective. However, the relevance of these mechanisms in paediatric patients with SE remains unknown. To test this hypothesis, we combine clinical EEG recordings with dynamic causal modelling (DCM). This approach allows model-based inference of cortical synaptic coupling parameters based on EEG recorded across distinct oscillatory states. Our DCM revealed that dynamic changes in inhibitory synaptic coupling explain differences in EEG power spectra associated with BZP-treatment responsiveness, and guide the transition from ictal to interictal state. Furthermore, in silico simulations demonstrate that there are alternative routes to seizure termination even in cortical circuit models unresponsive to the BZPs. Together, our findings confirm that alterations in synaptic inhibition underlie BZP response during paediatric SE. More broadly this work further demonstrates the utility of computational modelling to validate insights from basic science in clinically accessible recordings in neurological disorders characterised by abnormal brain states.