FDA's AI-Enabled Medical Device List, updated on September 4, 2026, names 1,614 authorized devices, with decisions through June 29, 2026. Of these, 1,553 (96.2%) were cleared through 510(k), 40 were granted through De Novo, and 21 were approved through premarket approval (PMA).
Key takeaways
- FDA's list, updated on September 4, 2026, names 1,614 AI-enabled devices, with the latest decision dated June 29, 2026.
- Radiology accounts for 76.2% of the list, and radiology, cardiovascular and neurology devices together make up 90.3%.
- FDA authorized 181 AI-enabled devices in 2026 through June 29, against 173 in the same period of 2025.
- We ran every AI 510(k) through openFDA: 54.3% belong to companies outside the US, and South Korea ranks fifth, with 80.
- AI 510(k)s decided in 2025 took a median of 141 calendar days from FDA's receipt to its decision, and Special 510(k)s took 34.
Tracker status
| Figure | Current value |
|---|---|
| AI-enabled devices on FDA's list | 1,614 |
| FDA list page last updated | September 4, 2026 |
| Latest decision date in the list | June 29, 2026 |
| Our last check | September 26, 2026 |
How many AI medical devices has FDA authorized?
1,614, according to FDA's own list [1, 2]. FDA says the list identifies AI-enabled medical devices that are authorized for marketing in the United States [1].
Most were cleared, but not all. FDA's list calls them authorized [1], because each pathway has its own term:
| Pathway | Devices | Share | FDA's term |
|---|---|---|---|
| 510(k) | 1,553 | 96.2% | Cleared |
| De Novo | 40 | 2.5% | Granted |
| PMA | 21 | 1.3% | Approved |
Source: our analysis of FDA's AI-Enabled Medical Device List, downloaded September 26, 2026 [2].
The list runs behind the market. FDA updates it periodically [1]. FDA says devices authorized but without a published decision summary during its data collection period will be added in a later update [1]. So the true number authorized by today is higher than 1,614.
How fast is the list growing?
It has grown every year since 2015, though unevenly. Devices on the list, by year of FDA's decision [2]:
| Year | AI-enabled devices |
|---|---|
| 2018 | 65 |
| 2019 | 80 |
| 2020 | 114 |
| 2021 | 129 |
| 2022 | 162 |
| 2023 | 226 |
| 2024 | 235 |
| 2025 | 335 |
| 2026, through June 29 | 181 |
2024 added only 9 devices more than 2023, then 2025 jumped by 100, to 335, the largest year so far [2]. In 2026, FDA reached 181 by June 29, against 173 in the same period of 2025 [2]. Because the list lags, the 2026 figure may grow when FDA next updates it [1].
How long does FDA take to clear an AI device?
About 4.5 to 5 months, at the median. We measured the calendar days from FDA's receipt of each AI 510(k) to its decision, using FDA's 510(k) data [2, 3]:
| Decided in | AI 510(k)s | Median days | Middle half of cases |
|---|---|---|---|
| 2023 | 219 | 140 | 89 to 226 |
| 2024 | 225 | 147 | 100 to 206 |
| 2025 | 329 | 141 | 94 to 210 |
| 2026, through June 29 | 176 | 144 | 105 to 229 |
These are calendar days, not FDA review days. They include any time a submission waited on hold for the manufacturer's answers, so they are longer than FDA's own review clock.
Special 510(k)s are much faster. Manufacturers use a Special 510(k) for certain changes to their own cleared device [9]. Across all years, the 182 AI Special 510(k)s had a median of 29 days, against 146 for the 1,358 Traditional 510(k)s. For 2025 decisions, the medians were 34 and 153 days [2, 3].
Our priority markets, for AI 510(k)s decided from 2024 through June 2026: South Korea 147 days (43 devices), Japan 128 days (26) and Taiwan 196 days (14) [2, 3]. India had too few to compare.
Which specialties dominate?
Radiology, by far. FDA's list assigns each device a lead review panel [2]:
| Lead panel | Devices | Share |
|---|---|---|
| Radiology | 1,230 | 76.2% |
| Cardiovascular | 154 | 9.5% |
| Neurology | 73 | 4.5% |
| Anesthesiology | 30 | 1.9% |
| Gastroenterology and urology | 27 | 1.7% |
| Hematology | 22 | 1.4% |
| All other panels | 78 | 4.8% |
Six product codes cover 54% of the list. The most common primary product codes are all radiology devices [2, 4]:
| Product code | Device type | Devices |
|---|---|---|
| QIH | Automated radiological image processing software | 305 |
| LLZ | Radiological image processing system | 161 |
| IYN | Pulsed Doppler ultrasound imaging system | 124 |
| JAK | Computed tomography X-ray system | 103 |
| LNH | Magnetic resonance imaging system | 100 |
| QAS | Radiological computer-assisted triage and notification software | 86 |
QIH and LLZ fall under 21 CFR 892.2050, medical image management and processing systems [5]. QAS falls under 21 CFR 892.2080, radiological computer aided triage and notification software [5].
Which countries do AI devices come from?
More than half come from outside the US. FDA's list does not give a country, so we matched all 1,553 AI 510(k)s to the applicant's country in FDA's 510(k) data [2, 3]. Companies from 35 countries hold them:
| Country | AI 510(k)s |
|---|---|
| United States | 710 |
| Israel | 121 |
| France | 95 |
| China | 84 |
| South Korea | 80 |
| Japan | 63 |
| Canada | 57 |
| United Kingdom | 48 |
| Netherlands | 43 |
| Germany | 33 |
| Taiwan | 30 |
Source: our analysis of FDA's AI list and openFDA 510(k) data, September 26, 2026 [2, 3]. De Novo and PMA devices are not included.
The largest holders on FDA's list are Siemens Medical Solutions USA (56 devices), Canon Medical Systems (42), Aidoc Medical (30) and Shanghai United Imaging Healthcare (29) [2].
The non-US share is steady. Companies outside the US held between 57% and 59% of each year's AI 510(k)s from 2023 through June 2026 [2, 3].
Our priority markets:
| Market | AI 510(k)s | Largest holders |
|---|---|---|
| South Korea | 80 | Samsung Medison (13); JLK, AIRS Medical and Coreline Soft (4 each) |
| Japan | 63 | Canon Medical Systems (41) |
| Taiwan | 30 | Ever Fortune.AI (8) |
| India | 13 | Qure.ai (9) |
| Turkey | 2 | Smart Alfa Teknoloji (2) |
| Malaysia, UAE, Saudi Arabia | 0 | None |
South Korea's count has grown the most of these markets. It had 3 AI 510(k)s in 2020, 17 in 2024 and 14 in 2025, and already 12 in the first half of 2026 [2, 3].
What do the newest novel AI devices look like?
Novel AI devices, with no suitable predicate, go through De Novo. Four were granted in 2026 through June 29, and three of them are outside radiology [2]:
| Date granted | Device | Company | Lead panel |
|---|---|---|---|
| May 22, 2026 | PCWP Analysis Software | Cardiosense | Cardiovascular |
| May 21, 2026 | DeepView AI System | Spectral MD | General and plastic surgery |
| March 17, 2026 | Tyto Insights for Eardrum Bulging Detection | Tyto Care | Ear, nose and throat |
| February 11, 2026 | Delivery Date AI | Ultrasound AI | Radiology |
A device classified through De Novo may then be used as a predicate for future 510(k)s, when applicable [8]. Our SaMD and AI learning path covers which pathway fits software and AI devices.
How many AI devices have a PCCP?
We cannot tell from FDA's list, but at least two. A PCCP lets a manufacturer make planned changes to an AI device without a new submission for each one [6]. Only 2 device names on FDA's list mention one: Tyto Insights for Wheeze Detection (with PCCP), cleared on June 2, 2026, and Fibresolve (with PCCP), cleared on November 7, 2025 [2].
Treat this as a floor. The list does not record PCCPs, so devices with a PCCP that do not say so in their name are not counted. FDA's final PCCP guidance for AI-enabled device software functions was issued on December 4, 2024, and updated on August 18, 2025 [6]. Our guide to whether a PCCP is only for AI devices explains who can use one.
What does FDA's list not tell you?
- It lags. Authorized devices appear only after FDA publishes their decision summaries [1].
- It has no country, PCCP or model-type fields. Our country and PCCP figures come from matching and from device names [2, 3].
- It does not yet flag large language models. FDA says it will explore ways to identify devices that use foundation models, including large language models [1].
- FDA's lifecycle guidance for AI devices is still a draft. Its lifecycle guidance for AI-enabled device software functions was issued as a draft on January 7, 2025 [7].
Our guide to FDA regulation of AI medical devices covers what FDA expects in an AI submission.
How do we update this tracker?
Every month, we download FDA's list, recompute each figure on this page, and match any new 510(k)s to their countries and review dates [2, 3]. When FDA updates its list, we add the new devices and record the change below.
| Our check | FDA list update | Devices | Change |
|---|---|---|---|
| September 26, 2026 | September 4, 2026 | 1,614 | Tracker launched |
What does this mean for your AI device?
In radiology, expect a crowded field. With 1,230 radiology devices on the list, you have many possible predicates, and many competitors [2]. Outside radiology, you may find fewer predicates, and a De Novo may be the right route.
Plan for change early. If you expect to retrain or update your model, a PCCP in your first submission can save later 510(k)s [6]. The guiding principles for good machine learning practice are covered in our GMLP guide.
Complizen's predicate research ranks possible predicates by similarity in intended use, technology and patient population, and includes De Novo precedent search.
Frequently asked questions
How many AI medical devices has FDA cleared?
FDA's AI-Enabled Medical Device List names 1,614 authorized devices, with decisions through June 29, 2026. Of these, 1,553 were cleared through 510(k). The other 61 were granted through De Novo or approved through PMA, so "authorized" is the accurate word for the total.
How often does FDA update its AI device list?
FDA says it updates the list periodically, not on a fixed schedule. The current version was posted on September 4, 2026, with decisions through June 29, 2026. Devices whose decision summaries were not yet published are added in a later update.
Which specialty has the most AI devices?
Radiology, with 1,230 of the 1,614 devices, or 76.2%. Cardiovascular devices are second, with 154 (9.5%), and neurology third, with 73 (4.5%). Together these three specialties account for 90.3% of FDA's AI-enabled device list. The top product code, QIH, is automated radiological image processing software, with 305 devices.
What share of AI devices use the 510(k) pathway?
96.2%. Of the 1,614 devices on FDA's list, 1,553 were cleared through 510(k), 40 (2.5%) were granted through De Novo and 21 (1.3%) were approved through premarket approval. Most AI devices reach the market by showing substantial equivalence to a predicate.
How many AI devices were authorized in 2025?
FDA's list includes 335 AI-enabled devices with decisions in 2025, the largest year so far. In 2024 the total was 235. In 2026 through June 29, FDA had authorized 181, against 173 in the same period of 2025. The count has risen every year since 2015.
How many AI devices come from outside the US?
We matched all 1,553 AI 510(k)s to the applicant's country: 843, or 54.3%, belong to companies outside the US, from 34 other countries. Israel leads with 121, followed by France with 95, China with 84 and South Korea with 80.
How many Korean AI medical devices has FDA cleared?
80, based on our match of FDA's AI list to 510(k) applicant countries. Samsung Medison holds the most, with 13. Korea had 17 AI 510(k)s in 2024 and 14 in 2025, and already 12 in the first half of 2026.
Does FDA's list show which devices use large language models?
Not yet. FDA says it will explore ways to identify devices that incorporate foundation models, including large language models, and encourages sponsors to describe such functions in their public summaries. The current list has no field for model type. For now, check each device's public summary.
How many AI devices have a PCCP?
FDA's list does not record PCCPs, so no exact count exists. Only 2 device names on the list mention one: Tyto Insights for Wheeze Detection and Fibresolve. The real number may be higher, because a device with a PCCP need not mention it in its name.
Is FDA's AI device guidance final?
Partly. FDA's guidance on PCCPs for AI-enabled device software functions is final, issued in December 2024 and updated in August 2025. Its broader lifecycle guidance for AI-enabled device software functions was issued as a draft on January 7, 2025. A draft guidance is not for implementation until FDA finalizes it.
Why is the number on this page different from other sources?
Other sources may use an older version of FDA's list, count only cleared devices, or add devices FDA has not yet listed. This page uses FDA's current list, downloaded September 26, 2026, and states the date of its latest decision.
How long does FDA take to clear an AI device?
For AI 510(k)s decided in 2025, the median was 141 calendar days from FDA's receipt to its decision. That count includes time on hold waiting for the manufacturer's answers. Special 510(k)s decided in 2025, used for certain changes to a cleared device, had a median of 34 days.
References
- FDA. Artificial Intelligence-Enabled Medical Devices. Content current as of September 4, 2026.
- FDA. AI-Enabled Medical Device List, CSV file. Downloaded September 26, 2026.
- openFDA. Device 510(k) API. Accessed September 26, 2026.
- openFDA. Device classification API. Accessed September 26, 2026.
- eCFR. 21 CFR Part 892, Radiology Devices, sections 892.2050 and 892.2080. Accessed September 26, 2026.
- FDA. Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions. Guidance issued August 18, 2025; originally issued December 4, 2024.
- FDA. Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations. Draft guidance issued January 7, 2025.
- FDA. De Novo Classification Request. Content current as of September 30, 2025.
- FDA. The Special 510(k) Program. Guidance issued September 13, 2019.