In the EU, it is often said that the AI Act overregulates artificial intelligence. The U.S. Food and Drug Administration (FDA) is also keeping a close eye on AI in medical devices. Rather than relying on a separate AI law, it draws on existing medical device regulations and guidelines. A recent Warning Letter shows how strictly the FDA interprets them for AI in medical devices. Through such public letters, the FDA notifies manufacturers of significant violations and requires a response within 15 business days.
A manufacturer had changed its FDA-cleared software in three ways: It added a new machine learning algorithm, switched from a desktop application to a web-based architecture and migrated the technology platform. According to the FDA, the new algorithm alone would have required a new premarket notification (510(k)). The rationale: A new model may perform differently and fail differently than the cleared version in certain patient groups, devices or clinical settings. This applies even if the update is meant to be an improvement. The FDA sees a risk of missed or inadequate diagnoses and demands that distribution of the new version be stopped.