The rapid adoption of wearable devices has significantly expanded the potential applications of EEG signals in biometric identification, including brain fingerprint recognition. EEG, being non-invasive and offering real-time data acquisition, provides an attractive solution for secure authentication systems. However, a major obstacle in implementing EEG-based brain fingerprint recognition in real-world scenarios is the inherent variability present in cross-session EEG data. This variability, arising from factors such as changes in user state, environmental influences, and equipment variations, severely impacts the reliability and stability of models over time. In addition, the heterogeneity of task-specific protocols further complicates the generalization of models, making it difficult to ensure consistent performance across different sessions or individuals. Given the increasing demand for secure, long-term authentication systems, overcoming these challenges is essential for the practical deployment of EEG-based brain fingerprint recognition technologies. This innovative framework is specifically designed to enhance cross-session EEG-based brain fingerprint identification under varying task protocols. The JDR-DAT framework tackles the problem of variability by disentangling identity-related features using mutual information estimation, ensuring that the model can reliably extract unique user information irrespective of session-related changes. Furthermore, by incorporating domain adversarial training, the framework improves robustness across different sessions, enhancing the model’s ability to maintain high accuracy over time. Extensive tests on longitudinal EEG data from two publicly accessible datasets, Rapid Serial Visual Presentation and Motor Imagery, validate the effectiveness of JDR-DAT, achieving average accuracies of 85.83 and 96.72%, respectively.