JMIR Publications

ABSTRACT

Clinical artificial intelligence (AI) tools increasingly draw on records held by more than one institution. Latin American countries, from Uruguay's national platform to recent laws in Colombia, Brazil, Argentina, Chile, and Mexico, are building the channels for this exchange. The provisions we examined address record custody, exchange, validation, and correction, including duties toward recipients of corrected data, but we did not identify one that designates who resolves a clinically relevant discrepancy between two institutions' records. Such discrepancies are common at transitions of care and can harm patients, and the closest international arrangements assign only part of that responsibility. This Viewpoint argues that where a clinical AI tool draws on records from more than one institution, local authorization should require an explicit agreement on who resolves clinically relevant discrepancies between those records, and on how the tool behaves pending resolution, even where each institution is meeting its existing duties to exchange, correct, and rectify the data in its charge. We explain what assigning that responsibility means: naming in advance who resolves the conflict, how quickly, and how the tool behaves meanwhile. We place it within an author-proposed framework compared with existing guidance, weigh its costs and the strongest objections, and propose next steps. The framework is not validated, no patient data were analyzed, and we do not claim that requiring the agreement improves implementation, safety, equity, or sustainability, of health care ad health systems, which requires evaluation.