AI image generation models, including advanced systems such as MidJourney, Canva, and DALL \(\cdot \) E 3, exhibit significant sensitivity to noise, which can disrupt the classification ability of Machine Learning (ML) models to classify generated images accurately. Even when models confidently predict the correct class, the introduction of noise can obscure critical patterns, leading to misclassifications. This vulnerability highlights the need for robustness and noise tolerance in ML models. To address this issue, we investigate the integration of persistent entropy and persistent homology into the classification process of AI-generated images. Persistent entropy quantifies the uncertainty of topological features, while persistent homology captures the multiscale structural characteristics of the data. Our experiments demonstrate that incorporating these topological data analysis techniques enhances the robustness and accuracy of ML models, even under noisy conditions. This study underscores the potential of persistent entropy and persistent homology to improve the reliability of AI-driven image classification systems, ensuring better performance in real-world applications where noise is prevalent.