Moral inference under stress: When we doubt our beliefs about good people

Stress shapes how we perceive and interpret the world and may also influence how we evaluate other people. Do we learn differently about others' moral character under stress? Recent computational models suggest that moral inference relies on Bayesian learning, where beliefs about others are updated in proportion to their uncertainty. We hypothesized that acute stress increases belief volatility - reflecting heightened uncertainty and flexible updating. In this preregistered study, participants (n = 122; 50% female) underwent the Trier Social Stress Test (TSST) or a control manipulation before completing a moral inference task, in which they repeatedly predicted and observed moral decisions of two agents with differing moral characters. Acute stress increased belief volatility specifically for good agents, but not for bad agents. The effect emerged only when good agents were observed first - that is, when cortisol reactivity peaked. Moreover, individual differences in cortisol responses predicted the magnitude of the volatility increase across groups. This provides converging evidence that hypothalamic-pituitary-adrenal (HPA) axis activation might play a key role in moral inference under stress, although cortisol-specific effects cannot be dissociated from the broader stress manipulation. Stress-related increases in belief volatility cannot be explained by shifts in participants' own moral preferences, as acute stress had no impact on moral decision-making in an independent task. Participants' harm aversion was instead predicted by chronic stress. Our findings reveal a cognitive-computational mechanism through which stress shifts moral inference - an effect that may aid coping with uncertainty but also contribute to social dysfunction in stress-related psychopathologies.