Abstract
Over the past five years, perinatal risk prediction using artificial intelligence has expanded rapidly, drawing on routine clinical records, ultrasound findings, and continuous physiologic signals to generate dynamic high-risk scores across pregnancy. These tools promise earlier identification of complications, more precise monitoring, and better targeting of preventive resources, but their net benefit will hinge on how risk labels shape care and lived experience. In this Perspective, we conducted a targeted, non-systematic narrative synthesis integrating (i) evidence on AI-based obstetric risk prediction, (ii) lessons from prenatal screening and high-risk labeling, and (iii) principles and guidance on trustworthy digital health, equity/fairness, risk communication, and reproductive-data governance to examine how probabilistic outputs can unintentionally increase distress and inequity. We argue that risk labeling may fuel predictive anxiety when probabilities are interpreted deterministically, and secondary anxiety when intensified surveillance is experienced as confirmation of danger. We also outline discrimination pathways, including biased data and labels that over-flag socially disadvantaged groups, defensive clinical escalation that drives over-medicalization, and social or employment harms when sensitive pregnancy data are reused beyond care. To balance benefit and harm, we propose integrated safeguards: transparent model documentation, local and subgroup calibration, continuous fairness monitoring, structured and patient-centered risk communication with meaningful choice, strict privacy and purpose-limitation protections, and tiered psychological support embedded in clinical pathways. Future deployments should proceed as monitored pilots that jointly track clinical outcomes, equity, and perinatal mental health before scale-up.