Among the various disruptions to landscape sustainability caused by the full-scale invasion of Ukraine, the war-related impacts on the supply of cultural ecosystem services (CES) remain understudied. We employed a novel GeoAI framework to spatially assess landscape suitability for outdoor recreation in Eastern Ukraine (Luhansk, Donetsk, and Kharkiv oblasts). We trained a Random Forest model on 20,124 spatially unique outdoors visitation locations obtained from georeferenced records from Flickr, Twitter, and VKontakte made in 2015–2022, as a proxy for pre-war met recreational demand, with AlphaEarth satellite embeddings representing environmental conditions to predict environmental (not accounting for accessibility or safety) landscape suitability for recreation in the areas with no detected visitation. We achieved good predictive quality (mean spatially blocked cross-validation AUC = 0.888) in mapping baseline landscape suitability. By applying the same model to wartime satellite embeddings (2022–2024) and isolating war-related effects from climatic and seasonal noise, we detected war-related anomalies in landscape suitability. Our results reveal the presence of both anomaly directions: direct landscape damage (suitability decline, 2,834.25 km²), concentrated near known battlefield zones, and war-wilding (suitability increase, 10,736.75 km²), driven by land abandonment and spontaneous vegetation dynamics. Revealed discrepancies demonstrate a low-to-moderate Spearman’s correlation with war intensity, mapped from open sources, with grasslands exhibiting a correlation up to 0.46. This study is the first of its kind: a regional-scale, spatially explicit assessment of the loss of cultural ecosystem service supply under war pressures, offering a transferable methodology for monitoring landscape sustainability for outdoor recreation in other conflict areas worldwide.