The Food and Drug Administration published a final order in the Federal Register on Sept. 11, 2026, that adds a category it calls cardiovascular machine learning-based notification software to its rules for medical devices. FDA defines it as software that uses machine learning, a form of AI, to suggest how likely a patient is to have a heart or blood-vessel condition, so the patient can be referred or tested further. It is meant as a starting point for more testing, not a diagnosis, and it is not meant to detect irregular heartbeats.
The category itself dates from Aug. 3, 2023, when FDA granted Viz.ai's request to classify its Viz HCM software. Viz HCM checks standard 12-lead electrocardiograms (ECGs) from patients 18 and older for signs of hypertrophic cardiomyopathy, a condition in which the heart muscle becomes larger and thicker than normal. FDA says the classification has applied since that date; the new order, effective Sept. 11, 2026, writes it into the regulations.
Viz HCM is meant to flag patients for follow-up, not to replace the usual ways doctors diagnose the condition, and a negative result is not meant to rule out follow-up. It is not for patients with implanted pacemakers.
The order puts the category in Class II. FDA sorts medical devices into three classes by risk, from Class I, the lowest, to Class III, the highest, and Class II devices must meet specific safety requirements. For this category, makers must test their software on data from at least three geographically diverse sites, kept separate from the sites used to train it. They must also check whether users could misread its results, and their labels must warn users not to skip follow-up just because the software found nothing.
FDA has not exempted the category from review before sale, so makers of similar tools still need FDA clearance before selling them. FDA says that because Viz HCM was placed in Class II through the agency's process for new kinds of devices, later products of the same type can seek that clearance by comparing themselves with it, without filing their own new-device request or a full premarket approval application.
The order requires makers to report how well their software performs, such as how often it catches the condition and how often it raises false alarms, but it gives no such figures for any product.