Recently, it has been suggested that brain dysconnectivity in patients with schizophrenia might contribute to the wide-ranging cognitive deficits that characterize the disease. Graph theoretical analysis offers a unique method for studying how architectural alterations in large-scale brain networks may contribute to cognitive impairments in these patients. Implementing this technique, we analyzed the functional brain activity during a predictive switch-drift task from 22 patients with schizophrenia and 22 matched healthy controls. We specifically calculated task-based global graph measures for the functional networks that were activated during expected trials, trials requiring a flexible updating of predictions, and trials that required the stabilization of predictions. By implementing multivariate Bayesian generalized linear models, we found functional network alterations during all event types, which indicated less centralized, less integrated, and simultaneously less segregated network topology in patients with schizophrenia compared to controls. In addition, the rate of correctly detected switches, requiring flexible updating of internal models, predicted global graph measures differently for patients compared to controls. In particular, lower cognitive flexibility in patients was associated with stronger deficits regarding the integration of functional networks. Overall, the results suggest that the functional network topology in schizophrenia is less optimally organized compared to the neurotypical brain, which results in less flexible and stable cognitive processing.