Timely and accurate diagnosis of oral diseases is vital for effective dental care, yet conventional deep learning approaches face two critical challenges: patient data privacy and model interpretability. In this study, we propose SplitX-OralNet, a novel split learning framework integrated with explainable artificial intelligence (XAI) to address these limitations in the multi-class classification of oral diseases from intraoral photographic images. Our architecture employs a partitioned MobileNetV2 model, where feature extraction occurs on the client-side and classification is performed on the server-side, ensuring privacy-preserving computation without compromising diagnostic accuracy. We evaluate our method on two benchmark datasets, including Oral Diseases and Dental Diseases, achieving classification accuracies of 93.30% and 99.18%, respectively. Further evaluation using F1-score, ROC-AUC, and confusion matrices demonstrates the robustness and reliability of the proposed framework. To enhance transparency, we incorporate Grad-CAM and TCAV techniques, providing class-specific visual explanations that align with clinical insights, thereby enlightening the audience. The results highlight SplitX-OralNet’s potential as a scalable, trustworthy, and privacy-compliant AI solution for clinical deployment in dental diagnostics.