For decades, most medical devices have entered American hospitals not through rigorous independent testing, but through a regulatory shortcut that assumes similarity implies safety — an assumption the record of recalls quietly contradicts. Now, researchers from Indiana University, Harvard Kennedy School, and Emerging Health Consulting have proposed a machine-learning system that could help the FDA sort the genuinely safe from the genuinely dangerous, reserving human expertise for the cases where it matters most. The promise is substantial: fewer unsafe devices reaching patients, less burden on