Almost every claim about "how much AI is in healthcare" eventually points back to one artifact: the list of AI-enabled devices the US Food and Drug Administration has authorized for marketing. It is the closest thing the field has to a census. This page reads that list plainly — how fast it has grown, which specialties dominate it, and what the devices on it actually do — with each figure tied to the regulator's own record or to a peer-reviewed analysis of it. As of July 2026.
How fast is the cleared-device list growing?
The trajectory is steep. A peer-reviewed analysis drawing on the FDA database counted 692 AI/ML-enabled devices authorized by 2023 — roughly a 20-fold increase over the mean annual authorization rate between 1995 and 2015 2. Then 2024 set a single-year record with 168 machine-learning- enabled Class II authorizations 2. A separate taxonomy built a research cohort of 1,016 authorizations to classify what the devices do 3, and the FDA's living list has continued past that as new clearances land 1.
| Period | AI/ML-enabled authorizations | Note | Source |
|---|---|---|---|
| 1995–2015 | low single digits per year | the pre-boom baseline | 2 |
| Cumulative by 2023 | 692 | ~20× the 1995–2015 annual mean | 2 |
| 2024 (single year) | 168 | record annual total | 2 |
| Taxonomy research cohort | 1,016 | classified set of authorizations | 3 |
| Living FDA list | still growing | updated as clearances land | 1 |
Read the column carefully: a cumulative total and a single-year total answer different questions, and the taxonomy's 1,016 is a research cohort with its own cutoff — rather than a fourth data point on the same curve. They agree on direction, not on a single "true" number.
Which specialty owns the list?
The specialty split is the most durable fact here. Of the 168 devices authorized in 2024, 74.4% (125) were radiology — with cardiovascular a distant second and neurology third 2.
| Device panel | 2024 authorizations | Share |
|---|---|---|
| Radiology | 125 | 74.4% |
| Cardiovascular | 11 | 6.5% |
| Neurology | 10 | 6.0% |
| Anesthesiology | 5 | 3.0% |
| Gastroenterology–Urology | 5 | 3.0% |
| Dental | 3 | 1.8% |
This is no 2024 quirk. Radiology has been the largest panel every year the list has been tracked, for structural reasons: imaging produces standardized, digital, labelled data at scale, and the review pathway for image-analysis software is well worn. When people say "healthcare AI," the authorized reality is still, overwhelmingly, software that reads a scan.
What do the cleared devices actually do?
Counts say how many; the taxonomy says what kind. Across its 1,016 authorizations, quantitative image analysis was the most common function — though its relative share has started to decline as other uses appear. More than 100 devices use AI to generate data (for example, synthesizing or enhancing an image) rather than only measuring it, and — notably for a field dominated by headlines about chatbots — none of the authorized devices yet use large language models 3. The clinic-grade, cleared reality lags the consumer conversation by a wide margin.
The transparency gap
Approval volume has outrun disclosure. Among the 2024 authorizations, only 29.2% reported both sensitivity and specificity, just 15.5% included demographic data on the study population, and a Predetermined Change Control Plan — the mechanism that lets a model be updated after clearance — appeared in only 16.7% of summaries; cybersecurity considerations fared better at 54.2% 2. For anyone evaluating a device, the lesson is blunt: a clearance confirms the FDA's review threshold was met, not that a full performance-and- equity profile is published for you to read.
Where are the rules heading?
The FDA list is a US artifact, but the devices on it increasingly answer to more than one regulator. Under the EU AI Act (Regulation 2024/1689), AI that is a regulated medical device is treated as high-risk, and the obligations for those products apply from 2 August 2027 4. A device cleared in the US today will meet a second, stricter conformity regime to reach the EU market — a gap our global regulation tracker follows in detail.
How to read these numbers
Three cautions matter most. First, inclusion criteria move the total: the FDA's list is built from AI-related terms in authorization documents and is explicitly not comprehensive, so different credible sources report different counts for the same agency. Second, the as-of date is part of the number — a list that grows monthly means any total is a snapshot; each figure here carries the period it came from. Third, an authorization is a market-entry signal, not an outcome: it says a device cleared review, not that it improves care in your setting. For the effect side of the ledger, see our clinical trial results tracker and the cluster hub, AI in healthcare statistics.
Sources and method
The counts here come from the FDA's own AI-Enabled Medical Device List 1 and from two peer-reviewed analyses of it — a 2024 authorization study 2 and a taxonomy of 1,016 authorizations 3 — with the cross-border rule set from the EU AI Act text 4. We revisit this page on a 90-day cycle and whenever the FDA updates its list or a new analysis re-tallies it by year or panel.