Nature Medicine: on-prem AI agents match cloud on diagnoses

An open-weight AI model deployed on-premise in hospitals matched the diagnostic accuracy of leading cloud-based AI in detecting seven diseases, achieving 90.0% accuracy versus 90.7%. The research introduces a reliability signal, ConsistencyDx, which filters cases to defer uncertain diagnoses for human review, ensuring privacy and institutional control without sending data externally.

This Wire brief sits within Fusion42's coverage of Digital Health, AI Infrastructure and Healthtech Infrastructure.