Regulation

The FDA AI-enabled device list: a statistics tracker

A dated read of the FDA's AI-Enabled Medical Device List — how many devices carry an authorization, which specialties dominate, which pathways they take, and how thinly they report performance and demographics — each figure tied to the FDA or a peer-reviewed census. As of July 2026.

By Jonas WeirReviewed by Jonas Weir · editorial reviewUpdated

The short version

  • By the FDA's own January 6, 2025 count, the agency had authorized more than 1,000 AI-enabled devices through established premarket pathways; a peer-reviewed taxonomy reviewed 1,016 authorizations.
  • Growth is steep and recent: over 690 ML-enabled devices were authorized across 1995–2023, and 2024 alone added 168 — a record year.
  • Concentration is the defining fact: radiology accounted for 74.4% of 2024 authorizations, far ahead of cardiovascular (6.5%) and neurology (6.0%).
  • Transparency reporting stays thin: among 2024 authorizations, only 29.2% reported both sensitivity and specificity, 15.5% gave demographic data, and 16.7% carried a Predetermined Change Control Plan.
  • The list is a curated registry rather than a census of every AI tool in care — read the counts with the FDA's own inclusion caveats, and confirm any single device's status against the live list.

The FDA maintains a public AI-Enabled Medical Device List — a registry of the devices with an AI component that have cleared the agency's premarket review and are authorized for US marketing. It is the closest thing the field has to an official scoreboard, and it is quoted constantly, usually as a single round number stripped of its date. This page does the opposite: it holds the list at arm's length as a set of dated snapshots, tied to the FDA's own statements and to peer-reviewed censuses that have counted the same authorizations, and it keeps the caveats attached. As of July 2026.

The list at a glance

Each row below is a specific measurement with a specific cutoff. They do not all count the same universe — the FDA's "AI-enabled devices" figure and the peer-reviewed "ML-enabled Class II" tallies use different definitions — so read the rows as complementary readings, not as one running total.

Snapshot (as of)MetricValueSource
1995–2023, cumulativeML-enabled devices authorizedover 6903
Jan 6, 2025 (FDA statement)AI-enabled devices authorized, all pathwaysmore than 1,0001
2025 taxonomy censusauthorizations reviewed1,0162
2024 calendar yearnew ML-enabled Class II devices168 (record year)3
Through May 30, 2025devices carrying an authorized PCCP264

Two facts stand out immediately. First, the curve is steep and recent: it took nearly three decades to reach the low hundreds, and a single recent year added 168 devices 3. Second, the FDA's own framing is deliberately about process, not hype — the agency describes these as devices "authorized ... through established premarket pathways," meaning each one met the applicable premarket requirements for its risk class 1.

Cumulative growth: three decades, then a surge

The long tail matters. A cross-sectional analysis of the FDA's authorizations records that the agency "authorized over 690 machine learning (ML)-enabled medical devices between 1995 and 2023," and then "authorized 168 ML-enabled Class II devices in 2024" — described by the authors as a record year 3. The FDA's January 2025 statement that it had cleared "more than 1,000 AI-enabled devices" is consistent with those numbers once you account for the wider net the agency casts (all AI-enabled devices across pathways, rather than ML-enabled Class II clearances alone) 1.

A separate peer-reviewed taxonomy took a structural rather than a chronological view, reviewing "1016 FDA authorizations of AI/ML-enabled medical devices" to classify what the AI actually does 2. Its headline finding is worth holding onto as the list keeps growing: "Quantitative image analysis remains the most common application, but its relative proportion has declined recently" — the field is broadening beyond image measurement. The same analysis notes that "over 100 devices leverage AI for data generation, though none yet involve LLMs" 2. As of this writing, large language models are absent from the authorized-device list even as they dominate the wider conversation.

Specialty concentration: radiology is the list

If one number explains the shape of every AI-device tally, it is the radiology share. Among 2024 authorizations, the panels broke down as follows 3:

Clinical panelShare of 2024 authorizationsDevices
Radiology74.4%125
Cardiovascular6.5%11
Neurology6.0%10
All other panels combined~13.1%22

Roughly three of every four newly authorized ML-enabled devices in 2024 were radiology tools 3. This concentration is why a rising total device count tells you less about AI reaching the bedside broadly than it might seem: much of the growth is more imaging software, in a specialty that was already the densest. It also means the list under-represents specialties where AI is discussed heavily but authorized rarely.

What the AI actually does

Specialty is only one axis. The peer-reviewed taxonomy classified the 1,016 authorizations by function — what the AI does with the data — and found the field quietly widening. Quantitative image analysis remains the single most common application, but "its relative proportion has declined recently" as other uses grow 2. Notably, "over 100 devices leverage AI for data generation" — producing images or signals rather than only measuring them — "though none yet involve LLMs" 2. That last clause is the one to sit with: as of this writing, the large language models driving most of the public conversation about AI in clinical care have yet to appear as authorized devices on the list at all. The gap between what is discussed and what is cleared is wide, and this tracker is where you can watch it close.

Pathways and sponsorship

How devices reach the list is as revealing as how many arrive. The 2024 cohort concentrated almost entirely in the moderate-risk clearance route 3:

Attribute (2024 authorizations)Value
Cleared via 510(k)94.6%
Cleared via De Novo5.4%
Sponsored by non-US companies57.7%
Median FDA review time162 days (510(k) 151; De Novo 372)

The dominance of the 510(k) pathway — which clears a device by demonstrating substantial equivalence to an existing predicate — shapes the evidence you can expect: most authorized AI devices were cleared without a prospective trial to reach market. The internationalization is its own signal: a majority of 2024 sponsors were based outside the US 3, a reminder that this is a global development pipeline funneling through one national gate. Speed compounds the effect — the median review took 162 days, but that figure splits sharply by route, at 151 days for a 510(k) against 372 days for a De Novo 3, so the faster pathway is also the busier one.

Predicate lineage

Because most devices clear by equivalence, the list is increasingly built on its own recent history. Among the 159 devices cleared via 510(k) in 2024, "97.5% cited an identifiable predicate," the median predicate was 2.2 years old, and 64.5% of those predicates were themselves ML-enabled; outright predicate reuse stayed uncommon at 9.9% 3. The picture is of a field maturing on its own lineage — new AI devices cleared largely by reference to recent AI predicates. That compounds the speed of clearance, and it raises the stakes on scrutinising what each original predicate actually proved, since a thin proof at the root can propagate down a whole family of devices.

Transparency reporting: the thin column

The most sobering cut is what device summaries actually disclose. In the 2024 cohort 3:

Element reported in the device summaryShare of 2024 authorizations
Both sensitivity and specificity29.2%
Demographic data15.5%
A Predetermined Change Control Plan16.7%
Cybersecurity considerations54.2%

One column is filling faster than the rest. Cybersecurity considerations appeared in 54.2% of 2024 summaries 3 — the only element in the table a majority of devices addressed — reflecting a broader FDA push to treat a model and its data pipeline as an attack surface in their own right. It is the disclosure most likely to become standard first, and a useful contrast: where the FDA has pressed hard and consistently, reporting rises.

Fewer than a third of authorizations reported both sensitivity and specificity, and roughly one in six disclosed demographic data 3. The analysis concluded that although 2024 "marked a record year for ML-enabled device approvals and internationalization, uptake of PCCPs and transparent performance and demographic reporting remained limited." That gap is the specific problem the FDA's draft total-product-lifecycle guidance sets out to close by recommending model-card-style disclosure and evidence that a device benefits demographic groups similarly 5; expect the transparency column to be a live figure as that guidance is finalized and its recommendations flow into future summaries.

The PCCP flag

A newer entry in the list is the marker showing a device shipped with a Predetermined Change Control Plan — FDA-authorized permission to make specified future updates without a fresh submission. A cross-sectional analysis identified "26 AI/ML-enabled medical devices with authorized PCCPs" through May 30, 2025, of which "92% were cleared via the 510(k) pathway, and all were classified as moderate risk" 4. The most common authorized change was model retraining. Set against the 16.7% PCCP share of the single 2024 cohort 3, the picture is of an adoption curve that is real but still early — worth tracking as its own line, which we do in the companion guide on PCCPs in practice.

The direction of travel is the thing to watch. Each FDA list refresh adds more entries carrying a "(with PCCP)" marker, and as the total-product-lifecycle guidance is finalized, the share of summaries reporting a plan should climb from the 16.7% recorded in 2024 3. Tracked over time, that share is a cleaner signal of how the field is maturing than the raw device count, which — as the radiology concentration shows — a single dense specialty can inflate on its own. It is the reason this tracker keeps the PCCP line, the specialty mix, and the transparency columns as separate readings rather than collapsing them into one headline figure.

How to read these numbers

Five cautions travel with every figure here.

First, the list is curated, not exhaustive. The FDA states plainly that the list is not comprehensive and does not capture every AI-enabled device, and that entries were identified in large part from AI-related terms in marketing-authorization summaries; tools described without those terms can be missed. Read the counts as a well-kept floor rather than a full census.

Second, the definitions differ across sources. The FDA's "AI-enabled devices" tally, the "1,016 authorizations" taxonomy, and the "168 ML-enabled Class II devices in 2024" census draw slightly different boundaries 123. Compare rows within a source, not across them, and never subtract one source's number from another's.

Third, the counts are perishable. The FDA refreshes the list on its own cadence, and each refresh moves the total. Any headline number is only true as of its cutoff date — which is why this page is stamped and revisited every ninety days.

Fourth, authorization is not validation. A place on the list means a device cleared its premarket bar for its risk class; it does not certify real-world performance, and — as the transparency column shows — many summaries report little of the evidence a deploying clinician would want 3.

Fifth, this is orientation, not regulatory advice. Confirm any specific device's authorization status, clearance number, and indications directly against the live FDA list, and involve regulatory or compliance counsel before relying on a device's status for a purchasing or clinical decision.

Sources and method

The FDA's own count and framing come from its January 6, 2025 announcement 1. The structural taxonomy of 1,016 authorizations is drawn from a peer-reviewed analysis 2; the year-by-year, specialty, pathway, and transparency figures from a cross-sectional census of 2024 authorizations 3; the PCCP counts from a cross-sectional analysis of the FDA's public list 4; and the transparency expectations from the FDA's draft lifecycle guidance 5. Every figure is tied to the primary source cited beside it, and where two sources count different universes we say so. For the same data cut by year and specialty, see our tracker of FDA-cleared AI devices by year and specialty; for the wider regulatory picture, see the global AI in health regulation tracker. We revisit this page every ninety days and whenever the FDA refreshes its list or a new census is published.

Questions & answers

  • How many AI-enabled devices has the FDA authorized?

    As of its January 6, 2025 announcement, the FDA stated it had authorized more than 1,000 AI-enabled devices through established premarket pathways, and a 2025 peer-reviewed taxonomy reviewed 1,016 authorizations. The exact running total changes with each list refresh, so treat any single figure as a dated snapshot and confirm it against the live FDA list.

  • Which medical specialty has the most FDA-authorized AI devices?

    Radiology, by a wide margin. Among devices authorized in 2024, radiology accounted for 74.4%, followed by cardiovascular at 6.5% and neurology at 6.0%. This roughly three-quarters radiology share has held for years and shapes almost every headline count.

  • Do FDA-authorized AI devices report their accuracy and demographics?

    Often not fully. In a cross-sectional analysis of 2024 authorizations, only 29.2% reported both sensitivity and specificity and 15.5% provided demographic data. The gap is part of why the FDA's draft lifecycle guidance presses harder on transparency.

Sources

  1. US Food and Drug Administration. FDA Issues Comprehensive Draft Guidance for Developers of Artificial Intelligence-Enabled Medical Devices (press announcement). January 6, 2025. www.fda.gov/news-events/press-announcements/fda-issues-comprehensive-draft-guidance-developers-artificial-intelligence-enabled-medical-devices
  2. Singh R, Bapna M, Diab A, Ruiz E, Lotter W. How AI is used in FDA-authorized medical devices: a taxonomy across 1,016 authorizations. npj Digital Medicine. 2025;8:388. doi.org/10.1038/s41746-025-01800-1
  3. Machine Learning-Enabled Medical Devices Authorized by the US Food and Drug Administration in 2024: Regulatory Characteristics, Predicate Lineage, and Transparency Reporting. Biomedicines. 2025;13(12):3005. doi.org/10.3390/biomedicines13123005
  4. Regulating Flexibility for Artificial Intelligence: FDA Experience with Predetermined Change Control Plans. medRxiv preprint. 2025. doi.org/10.1101/2025.08.26.25334477
  5. US Food and Drug Administration. Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (Draft Guidance, issued January 2025). www.fda.gov/media/184856/download