Study: AI for maternal health must be inspectable, not just trustworthy

A new paper presented at the 2026 ACM Conference on Fairness, Accountability, and Transparency (FAccT) tackles a critical question: How should AI-powered information tools earn trust in high-stakes maternal health contexts? Led by Vaibhav Balloli and colleagues, the researchers conducted four synchronous focus groups with 24 participants across three stakeholder groups—birthing people, clinicians, and health workers (doulas, social workers, community health workers). The study addressed the United States' alarming ranking in preventable maternal deaths and stark racial disparities, examining how current AI systems often under-specify the socio-technical governance structures needed for safe use.

The inductive analysis revealed a central finding: in environments shaped by historical inequities, trustworthiness must be inspectable rather than asserted. While stakeholder groups diverged on what makes specific information credible, they converged on the need for transparency, recourse, and ecosystem complementarity. The authors identified four governance requirements: (1) support for social and identity-based sensemaking, (2) pluralistic verification practices, (3) inspectable governance with recourse mechanisms, and (4) ecosystem-aware integration that avoids shifting burden to patients. These findings challenge developers to move beyond simple factual accuracy and build mistrust-aware systems with transparent governance.