Who verifies and who continues? Personality and physiological predictors of anomaly-responsive oversight in AI-supported navigation

Intelligent systems increasingly place humans in oversight roles, yet users exposed to the same anomaly warning do not respond uniformly. Across two studies using an AI-supported navigation task, we examined whether warning design, focal HEXACO traits and trial-level physiology explained anomaly-specific evaluation and verification/override behaviour. Both studies used a 2 × 2 manipulation of warning salience and explanation style. Study 1 combined eye tracking and pupillometry; Study 2 used electrodermal activity. Prominent and especially explanatory warnings produced the largest behavioural and evaluative effects. Personality showed smaller effects, with behavioural increments sensitive to model specification; Conscientiousness was most consistently associated with verification, whereas Emotionality was associated with evaluative and physiological responsiveness. Repeated participant-level validation showed substantial incremental prediction from pupil features and smaller gains from electrodermal features, including under temporally restricted sensitivity analyses. The findings support a differentiated person-centred account of anomaly-responsive oversight while separating subjective trust from behavioural reliance and verification.