The hidden FDA problem with AI medical devices is not approval — it’s what happens after
STAT reports that AI medical devices have a ‘dirty FDA secret,’ pointing to the gap between clearance and real-world performance. The story suggests that regulation may be strongest at the moment of approval and weakest once systems are deployed, updated, or used in new settings. That gap is where many of the most important safety questions now live.
The most consequential issue in AI medical devices may not be getting them through the FDA. It may be what happens after they are cleared. If models change over time, encounter new populations, or are deployed in workflows different from those used in validation, the original approval snapshot can quickly become outdated.
That is the “dirty secret” at the center of the problem: clearance is not the same as continuous safety. Traditional regulatory thinking often assumes a relatively stable device, but AI systems can drift, be retrained, or behave differently across institutions. The post-market phase is therefore not a footnote — it is the main event.
This creates a mismatch between regulatory process and technological behavior. The FDA can review evidence at a point in time, but healthcare organizations need assurance that performance remains acceptable in the wild. That means monitoring, audit trails, and update governance become as important as premarket submissions.
The article underscores a broader truth about AI in medicine: a cleared device is not a finished product. It is the start of an ongoing responsibility chain that includes vendors, regulators, and health systems. If that chain is weak, approval becomes a label rather than a guarantee.