Sometime in January 2026, the FDA’s running list of authorized artificial intelligence and machine learning-enabled medical devices quietly crossed the 1,000-device threshold and kept climbing, reaching roughly 1,357 total entries. It’s a number that has grown almost tenfold in less than a decade, and the composition of that list tells a clear story about where AI has actually taken root in American medicine—and where it hasn’t. Independent trackers monitoring the same FDA database found it had grown even further by March 2026, to somewhere between 1,451 and 1,524 entries depending on how duplicate or updated submissions are counted—a reminder that even the FDA itself cautions its published roster is “not a comprehensive resource of AI-enabled medical devices,” but rather a list built from AI-related terminology found in public authorization summaries, meaning the true count of AI-enabled tools in clinical use is likely higher still.
Radiology Dominates, By a Lot
Of the 1,357 authorized devices, about 1,039—77 percent—fall under radiology. That concentration isn’t an accident. Medical imaging produces enormous, well-labeled datasets (millions of X-rays, CT scans, and MRIs with known outcomes), which is exactly the kind of data modern deep-learning models need to train on. It’s also a domain where the AI’s job is relatively bounded: look at pixels, flag a pattern, hand the decision back to a radiologist. That combination—abundant data plus a narrow, well-defined task—has made imaging the low-hanging fruit of medical AI regulation for years, and 2026’s numbers show that trend accelerating rather than leveling off.
What’s Conspicuously Underrepresented
Compare that 77 percent concentration to fields like primary care decision support, mental health, or chronic disease management, where authorized AI devices remain a small fraction of the list. Those areas involve messier, less standardized data—clinical notes, patient-reported symptoms, longitudinal behavior—that’s harder to build a defensible, testable AI product around. It’s also why the clearance of UpDoc’s patient-facing large language model device in 2025 drew so much attention: it represented a rare foray outside the imaging-heavy status quo.
The Deployment Gap
Clearance is not the same as use. Healthcare organizations widely report that successful AI pilots often stall after regulatory approval, with a Carta Healthcare survey finding that EHR integration has now overtaken clinician trust as the leading barrier to actually deploying cleared AI tools inside hospital workflows. In other words, hundreds of the 1,357 cleared devices may exist on paper as validated, legal-to-sell products while sitting largely unused in the average hospital because nobody has built the plumbing to connect them to a doctor’s daily screen. Vendors that have successfully closed that integration gap are increasingly consolidating multiple narrow tools into single platforms rather than selling one-off algorithms—Aidoc, for instance, won clearance in January 2026 for what it calls the first multi-condition AI triage solution for body CT, now running across health systems including Advocate Health, AdventHealth, Mount Sinai, and Yale New Haven Health, suggesting the market is starting to reward integration and breadth over sheer device count.
A Test of Real-World Outcomes, Not Just Clearance
Regulators themselves seem aware that raw clearance counts are becoming a less meaningful measure of progress. The FDA’s February 2026 Technology-Enabled Meaningful Patient Outcomes (TEMPO) pilot program shifts scrutiny toward demonstrating measurable clinical benefit after deployment, not just technical accuracy during premarket testing. Industry commentary describes the healthcare AI sector as entering a “prove it or move aside” phase in 2026, after years where a device’s mere existence on the FDA list was treated as validation enough.
Money Is Forcing the Issue
Large payers and health systems have concrete financial stakes in making these numbers translate into real savings. UnitedHealth has projected close to $1 billion in AI-driven savings in 2026, and HCA Healthcare expects roughly $400 million, much of it from automating administrative and revenue-cycle tasks rather than clinical diagnosis—an indirect signal that even inside organizations flush with cleared AI tools, the easiest wins are still operational rather than diagnostic.
What’s Next
Expect the raw device count to keep climbing through 2026 as vendors continue to find narrow, well-defined imaging tasks to build around, but expect increasing pressure—from TEMPO and from hospital finance departments alike—to separate devices that are merely cleared from devices that are actually changing patient outcomes. The next milestone worth watching isn’t a bigger number on the FDA list; it’s how many of those 1,357 devices show up in a peer-reviewed study demonstrating they changed a real clinical result.
Photo: OpenDataInstitute / BY-SA via flickr
