This study aims to identify physiological features sensitive to human–automation trust (HAT) and to distinguish between high HAT (HHAT) and low HAT (LHAT) levels. A simulated aviation monitoring and takeover task was conducted under a 2 (reliability) × 2 (transparency) experimental design, in which HHAT and LHAT levels were defined based on their Jian scale scores. Paired t-tests were then employed to analyse HAT-sensitive features derived from five physiological modalities: electroencephalography (EEG), functional near-infrared spectroscopy (fNIRS), eye tracking, facial expression and electrocardiography (ECG). Finally, HAT discrimination models with varying interpretability were constructed using the identified sensitive features. The results indicated that: (1) compared with HHAT, LHAT exhibited significantly higher EEG 𝛼 relative power and lower 𝛿 relative power, shorter eye-tracking metrics including the average duration of whole fixations and the average and total durations of fixations in takeover and flight areas, as well as higher ECG metrics such as the standard deviation of normal-to-normal intervals, the low-frequency to high-frequency ratio, and the low-frequency ratio; (2) no significant differences were found in fNIRS oxygenated haemoglobin and deoxygenated haemoglobin concentrations or in facial micro-expression and action units across HAT levels and (3) the highly interpretable linear support vector machine model constructed based on HAT-sensitive features achieved optimal performance, yielding a fivefold cross-validation accuracy of 0.8065 and an F1-score of 0.7942. These findings can provide empirical evidence for identifying HAT-sensitive physiological markers and developing real-time HAT assessment methods, thereby supporting the monitoring of trust degradation that may compromise flight safety.