In medical settings, the real question is no longer which AI model looks strongest on paper, but whether the entire system can be governed, monitored, and safely changed after deployment.