Brain-computer interfaces (BCIs) have significant applications in medical treatments, rehabilitation, and various other fields. Electroencephalogram (EEG) signals are commonly used in motor imagery (MI) based BCIs, but they contain sensitive private information such as user identity, emotions, mental states, and other personal details. Without proper protection, unauthorized access to this private information is possible. To address these challenges, this study proposes an artificial intelligence (AI)-based data desensitization technique using format-preserving encryption (DD-FPE) to maintain user access control while optimizing MI task identification accuracy and minimizing information loss. The proposed approach involves three primary steps: signal pre-processing data using a Butterworth filter, extracting features based on Common Spatial Pattern (CSP) and DD-FPE, and applying a deep learning-based neural network (NN) algorithm for MI task detection. Our experiments on EEG-based BCI data achieved over 92% accuracy for all users or subjects, with computation times ranging from 2.37 to 2.89 s. This developed framework will guide future researchers in creating more effective AI frameworks in BCI technology, ensuring user control over access to their confidential information.