The FDA wants proof you did the AI homework before, not after
The part of the FDA's draft guidance that should worry sponsors is the timing. You have to show how you assessed a model's fitness for that specific use before you used it. From running AI governance reviews, I can tell you most teams can prove a tool works. Far fewer can produce the assessment they did before switching it on, because it never happened or it lives in someone's inbox. You can't backfill a pre-deployment decision once an inspector asks for it. If the FDA showed up tomorrow, which of your monitoring and data review tools has a dated record of why it was fit for purpose?
Source: FDA’s AI Risk-Based Guidance Shifts Validation Burden to Sponsors - The Clinical Trial Vanguard (The Clinical Trial Vanguard)