Author Summary Functional network disruption is a well-established finding in Alzheimer’s disease. Sensitive network-based biomarkers are however not available. We aimed to detect neuronal dysfunction at a predementia (mild cognitive impairment, MCI) stage of Alzheimer’s disease, by applying a network-level neural variability measure to magnetoencephalography data: the inverted joint permutation entropy (JPE inv ). This measure integrates information on local signal variability/complexity and nonlinear coupling. We found significant differences in JPE inv between subjects with subjective cognitive decline and MCI, primarily in the theta band. The diagnostic ability of the JPE inv was reported to be similar to that of relative theta power, the most potent neurophysiological biomarker of predementia Alzheimer’s disease to date. Abstract Increasing evidence suggests that measures of signal variability and complexity could present promising biomarkers for Alzheimer’s disease (AD). Earlier studies have however been limited to the characterization of local activity. Here, we investigate whether a network version of permutation entropy could serve as a novel biomarker for early-stage AD. Resting-state source-space magnetoencephalography was recorded in 18 subjects with subjective cognitive decline (SCD) and 18 subjects with mild cognitive impairment (MCI). Local activity was characterized by permutation entropy (PE). Network-level interactions were studied using the inverted joint permutation entropy (JPE inv ), corrected for volume conduction. The JPE inv showed a reduction of nonlinear connectivity in MCI subjects in the theta and alpha band. Local PE showed increased theta band entropy. Between-group differences were widespread across brain regions. Receiver operating characteristic (ROC) analysis of classification of MCI versus SCD subjects revealed that a logistic regression model trained on JPE inv features (78.4% [62.5–93.3%]) slightly outperformed PE (76.9% [60.3–93.4%]) and relative theta power–based models (76.9% [60.4–93.3%]). Classification performance of theta JPE inv was at least as good as the relative theta power benchmark. The JPE inv is therefore a potential biomarker for early-stage AD that should be explored in larger studies.