FDA wants AI devices watched for their whole life, not just at clearance

The FDA's first comprehensive draft guidance on AI-enabled device software asks manufacturers to manage these products across the total product lifecycle, including performance monitoring after deployment and Predetermined Change Control Plans that define how a model may change without a new submission. The signal for health systems: the agency now treats a cleared model as a starting point, not a finished product. Governance leads should expect to mirror that expectation internally, tracking whether a vendor's cleared tool still performs the way it did the day it was approved. As one analysis noted, AI bias can produce erroneous results in a systemic but unpredictable way. The comment period closed in April 2025; the direction of travel is set.