As high as 50% of Alzheimer’s disease patients experience “sundowning”, defined as an increase in severity of neuropsychiatric symptoms, including agitation, confusion, and anxiety selectively in the evening. While such symptoms influence the decision to institutionalize patients, few preclinical models for sundowning exist and its underlying neural mechanisms are unknown. Here, we establish a preclinical model of sundowning by phenotyping the sleep-wake activity cycle in control (Ctrl) and APP/PS1de9 (AD) mice across different ages in male and female mice. Next, we characterized their anxiety/exploratory behavior during sundown (ZT22-0) and sunrise (ZT12-14) using the elevated-plus maze (EPM) and open field (OF) assays. In AD but not Ctrl mice, we found degradation of sleep-wake cycle rhythmicity with age. In the EPM and OF assays, we found that older AD mice (24-month-old males and 12- and 24-month-old females) preferentially explore the open arms and display hyperlocomotion (24-month-old males and 24-month-old females) selectively at sundown, paralleling clinical symptoms such as “nighttime wandering” and motor agitation. Next, we combined the ArcCreERT2 x EYFP activity-dependent tagging strategy with a custom analysis pipeline to map ensembles differentially active at sundown and sunrise in AD and Ctrl mice. Network analysis revealed decreases in global connectivity measures such as degree and efficiency in AD mice at sundown. Overall, we have identified a striking behavioral abnormality in an AD model analogous to clinical features of sundowning. We have further mapped a corresponding neural activity network signature, revealing insights into the systems-level disruptions that may underlying this debilitating syndrome.