Monitoring of pilots’ mental workloads is crucial for flight safety. Given the scarcity of flight data, developing transferable mental workload detectors trained on accessible paradigms like the n-back task represents a critical advancement toward deployable neuroadaptive systems in aviation. In this work, we developed an LSTM framework that extracts task-invariant neural signatures from spectral power of EEG rhythms using a controlled n-back paradigm and transfers detection to flight simulations without retraining. The model achieved 79.25% ± 4.07% accuracy on n-back data, with hierarchical F1-scores revealing state-dependent efficacy: rest (0.858) > severe workload (0.773) > mild workload (0.750). When applied to professional pilots (n = 2) during flight scenarios of graded difficulty, workload detection ratios scaled increasing with perceived difficulty. This cross-task validity—validated despite limited flight labels—confirms that EEG workload markers transcend task boundaries. The proposed approach enables deployable neuroadaptive systems for real-time cognitive state monitoring.