Diabetes mellitus is one of the most global health challenges, needing early detection and effective management to reduce long‑term issues. Artificial intelligence (AI) techniques have shown promising capabilities in predicting diabetes using structured medical data, but most studies have focused on improving predictive accuracy while neglecting cybersecurity (CS) and patients’ data protection. This study suggests an integrated framework that combines machine learning(ML) with cybersecurity(CS) mechanisms. The proposed framework combines AES‑256‑GCM encryption, SHA‑256 integrity verification, and pseudonymization, along with anomaly detection using the Isolation Forest (IF)algorithm to improve data quality. The model was tested on Pima Indians Diabetes Dataset in three stages: before the integration of cybersecurity(CS), after the integration, after the removal of outliers. Results showed accuracy improving from 74.68% to 82.88%, with ROC‑AUC being stable at 0.82–0.84. The novelty of this study is adding a cybersecurity(CS) layer with a data quality improvement in a unified pipeline, strengthening model reliability. These facts show that achieving a balance between predictive accuracy, data quality, and cybersecurity(CS) is crucial for developing trustworthy AI‑based medical systems that will work for real‑world healthcare systems.