Why health AI needs protocols, not just pilots

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Health systems are rapidly adopting artificial intelligence (AI), with 2 in 3 physicians now using it for tasks like chart summaries and diagnoses—a 78% increase since 2023.1 Yet most clinical AI tools face limited US Food and Drug Administration (FDA) oversight, leading to inadequate evaluation.2 Over half of 1000 FDA-authorised AI devices reported no performance metrics and nearly half documented no clinical studies.3 Unlike pharmaceuticals, which require randomised trials and rigorous surveillance, AI tools are being deployed without comparable standards.