The Entropic Associative Memory (EAM) is a cognitively inspired computational model that combines probabilistic recognition with an efficient pattern storage and retrieval architecture. Previous evaluations, limited to small domains of about ten classes and a few thousand instances per class, showed promising results in classification and reconstruction. This work extends those studies in two ways. First, we assess scalability using Googles Quick, Draw! dataset, which includes over 300 classes and 100,000 instances per class. Second, we test EAMs rejection capability by storing only half of the classes and evaluating performance on the full dataset. Results confirm EAMs adaptability to large-scale scenarios and its effective rejection of novel stimuli, underscoring its potential as a robust and explainable AI model.