The Current State of FDA-Authorized AI-Enabled Medical Devices: 1,614 and Counting
AI in Healthcare • Medical Devices • FDA Regulation
Updated: September 23, 2026
The FDA's AI-enabled medical device ecosystem has passed the 1,600-device mark. The latest FDA database update contains more than 1,600 AI-enabled medical devices authorized for marketing in the United States. An analysis of the current dataset by Bertalan Meskó, MD, PhD, identified 1,614 entries, up from 1,451 in the previous year's dataset. (7, 8)
- What the 1,614-device figure actually means
- How rapidly AI medical devices are expanding
- Which medical specialties are using AI most
- FDA regulatory pathways explained
- Where are generative AI and LLM medical devices?
- What FDA authorization does — and does not — prove
- How FDA regulation is evolving in 2026
- What this means for patients and clinicians
- What it means for developers and health-tech companies
- Where AI medical devices are going next
- Frequently Asked Questions
- Sources and further reading
1. What Does 1,614 FDA-Authorized AI Medical Devices Actually Mean?
The FDA maintains an official AI-Enabled Medical Devices List identifying medical devices authorized for marketing in the United States that contain AI-enabled functions.
The FDA's current page states that over 1,600 AI-enabled medical devices had been authorized for marketing in the United States as of September 2026. The database includes devices spanning areas such as radiology, cardiovascular medicine, neurology, anesthesiology, gastroenterology and urology, hematology and other specialties.
The exact figure of 1,614 comes from the September 2026 analysis by Bertalan Meskó, MD, PhD, based on the newly updated FDA dataset. His analysis was published on September 22, 2026.
This distinction matters because the word "approval" has a specific regulatory meaning.
FDA authorization is not one single pathway
| FDA pathway | September 2026 entries | What it generally means |
|---|---|---|
| 510(k) | 1,553 | The device demonstrated substantial equivalence to a legally marketed predicate device. |
| De Novo | 40 | A pathway for certain novel low- to moderate-risk devices without a legally marketed predicate. |
| PMA | 21 | Premarket approval for higher-risk Class III devices, requiring FDA approval based on sufficient valid scientific evidence of safety and effectiveness. |
Accordingly, calling all 1,614 devices "FDA-approved" can create a misleading impression. The more precise description is FDA-authorized AI-enabled medical devices, with the specific regulatory pathway identified for each device.
See the FDA's explanations of 510(k) clearance, PMA, and its De Novo pathway for the formal regulatory definitions.
2. The Rapid Expansion of AI-Enabled Medical Devices
The growth of AI-enabled medical devices has accelerated dramatically over the last decade.
Meskó's September 2026 analysis reports the following annual counts from the FDA dataset:
| Year | AI-enabled device entries |
|---|---|
| 2016 | 18 |
| 2017 | 27 |
| 2018 | 65 |
| 2019 | 80 |
| 2020 | 114 |
| 2021 | 130 |
| 2022 | 163 |
| 2023 | 225 |
| 2024 | 235 |
| 2025 | 335 |
| 2026* | 181 |
*2026 represents the entries included in the September 2026 dataset update rather than a complete calendar year. Annual figures can change as the FDA revises, adds, removes or reclassifies database entries.
The important trend is not simply that there are more devices. The ecosystem is becoming more diverse, more clinically embedded and more closely connected to the software life cycle.
3. Which Medical Specialties Are Using AI the Most?
One feature of the current landscape is striking: radiology continues to dominate the FDA's AI-enabled device database.
In the September 2026 analysis, Meskó identified 1,230 radiology entries, compared with 154 cardiovascular entries and 76 neurology entries.
| Medical specialty/category | Entries in the September 2026 analysis | Approx. share of 1,614 |
|---|---|---|
| Radiology | 1,230 | 76.2% |
| Cardiovascular | 154 | 9.5% |
| Neurology | 76 | 4.7% |
| Anesthesiology | 30 | 1.9% |
| Gastroenterology-Urology | 27 | 1.7% |
| Hematology | 22 | 1.4% |
| Other categories | 75 | 4.6% |
Radiology's large lead is not especially surprising. Medical imaging produces highly structured, information-rich datasets that are well suited to machine-learning approaches for tasks such as detection, segmentation, quantification, reconstruction, prioritization and image enhancement.
The FDA's current examples include AI systems used for diagnostic imaging, skin-cancer-related imaging, diabetic-retinopathy detection, cardiac risk estimation and automated insulin dosing.
AI in medicine is much bigger than radiology
Radiology dominates the FDA database, but the underlying AI opportunity is broader. AI-enabled medical technologies can support:
- medical image interpretation and reconstruction
- signal and waveform analysis
- cardiovascular risk detection
- neurological assessment
- endoscopy and gastrointestinal applications
- pathology and tissue analysis
- clinical monitoring
- treatment planning
- drug-delivery or dosing functions
- workflow automation and decision support
The regulatory database therefore provides a useful window into where AI has reached the medical-device market — not necessarily a complete picture of where AI research or clinical experimentation is occurring.
4. FDA Regulatory Pathways: 510(k), De Novo and PMA
The regulatory pathway matters because two AI-enabled devices can both appear in the FDA database while having reached the market through substantially different regulatory routes.
510(k): the dominant pathway
According to the September 2026 analysis, 1,553 of the 1,614 entries were 510(k) submissions.
Under the 510(k) pathway, the manufacturer generally demonstrates that the new device is substantially equivalent to a legally marketed predicate device. The FDA explains that substantial equivalence involves comparison of intended use and technological characteristics, along with an assessment of safety and effectiveness information relevant to the differences.
This is why the presence of a 510(k)-cleared AI device should not automatically be interpreted as evidence that the underlying AI technology has undergone the same type of clinical evidence process associated with a PMA.
De Novo: novel devices without a predicate
The De Novo pathway is used for certain novel devices that are low to moderate risk and do not have an appropriate legally marketed predicate device.
De Novo can therefore be particularly relevant to emerging categories where a traditional predicate does not yet exist.
PMA: the highest-risk pathway
PMA is used for certain Class III devices and is the most stringent type of medical-device marketing application. FDA states that PMA approval requires sufficient valid scientific evidence to provide reasonable assurance of safety and effectiveness for the intended use.
5. Where Are Generative AI and Large Language Model Medical Devices?
This is one of the most interesting parts of the 2026 landscape.
Meskó's September 22, 2026 analysis stated that, as of September 2026, no device in the analyzed FDA database was identified as using generative AI or being powered by a large language model.
However, that statement needs an important qualification.
The FDA's current AI-device page says the agency is exploring methods to identify and tag medical devices that incorporate foundation models, including large language models and multimodal architectures. The agency also says sponsors are being encouraged to provide information that could help identify LLM-based functionality in future versions of the database.
In other words, the absence of an LLM-specific identification in the current FDA list should not be interpreted as proof that no regulated medical product anywhere uses generative AI functionality. It is more accurate to say that the current FDA AI-device database does not yet provide a mature, dedicated classification for foundation-model or LLM functionality.
The FDA is already preparing for the next phase
On August 18, 2026, the FDA published a discussion paper titled Considerations for the Regulation of Generative AI-Enabled Medical Devices. The document addresses questions around:
- risk assessment for generative-AI medical devices
- premarket evaluation
- postmarket monitoring
- regulatory approaches for systems whose behavior can be more difficult to characterize than conventional software
The discussion paper is explicitly not final guidance. It is intended to gather stakeholder feedback while the regulatory framework evolves.
6. What Does FDA Authorization Actually Prove?
This is arguably more important than the headline number.
An FDA-cleared or approved medical device has gone through the regulatory process applicable to its device classification and submission pathway. But the authorization should always be interpreted in the context of the device's intended use, indication, technology, pathway and evidence package.
FDA authorization does not automatically establish that:
- the technology improves every clinical outcome that matters to patients
- the device is superior to every alternative
- the algorithm works equally well across all populations
- the device is free of bias or performance limitations
- real-world performance will exactly match premarket testing
- the device can safely be used outside its authorized intended use
- an AI prediction is equivalent to a physician's diagnosis
This distinction is particularly important as AI systems become more complex and increasingly operate across different clinical environments.
Real-world performance is a separate question
AI medical-device performance can depend on factors such as patient population, imaging equipment, clinical workflow, data quality, prevalence of disease, deployment environment and changes to the software over time.
The FDA has increasingly emphasized life-cycle management rather than treating authorization as the end of the story.
7. How FDA Regulation Is Evolving in 2026
Predetermined Change Control Plans
The FDA's final guidance on Predetermined Change Control Plans (PCCPs) for AI-enabled device software functions provides a framework for manufacturers that expect certain planned AI modifications after authorization.
This is particularly relevant to machine-learning systems because software can potentially change over time as models are retrained, updated or improved.
The FDA guidance recommends describing planned modifications, the methodology for developing and validating them, and how their impact will be assessed.
See the FDA's 2025 final PCCP guidance.
Clinical decision-support software
In January 2026, the FDA issued final guidance on Clinical Decision Support Software, clarifying which software functions may fall outside the statutory definition of a medical device and which remain subject to FDA device policies.
This distinction matters because not every healthcare AI application is regulated as a medical device.
See the FDA's Clinical Decision Support Software guidance.
Generative AI regulation
The FDA's August 2026 discussion paper on generative-AI-enabled medical devices demonstrates that the agency is actively considering how conventional device regulation should adapt to technologies that may introduce new types of uncertainty, model behavior and postmarket risk.
That regulatory discussion is still developing.
8. What Does This Mean for Patients?
For patients, the growing number of AI-enabled medical devices is significant — but the number alone should not determine whether a patient should trust or choose a particular technology.
A more useful approach is to ask:
Patients should be particularly cautious about marketing language that turns a regulatory authorization into a much broader claim of clinical superiority, diagnostic certainty or guaranteed benefit.
9. What Does This Mean for AI Medical-Device Developers?
The rapid growth of FDA-listed AI-enabled medical devices creates both opportunity and regulatory complexity.
For developers, the central challenge is no longer simply proving that an algorithm can work under controlled conditions. Increasingly, questions concern the full product life cycle.
That includes:
- data quality and representativeness
- clinical validation
- human factors and usability
- transparency
- cybersecurity
- model performance monitoring
- change management
- postmarket surveillance
- bias and subgroup performance
- documentation of intended use and limitations
For AI systems that may change after deployment, the FDA's PCCP framework becomes particularly relevant.
10. What Comes Next for AI Medical Devices?
The first era of medical AI was dominated by relatively well-defined tasks such as image classification, detection, segmentation, reconstruction and signal analysis.
The next phase is likely to involve more sophisticated systems that combine multiple modalities and potentially multiple AI models.
Potential areas include:
- multimodal clinical decision support
- AI-assisted medical imaging across multiple modalities
- AI-enabled pathology and precision diagnostics
- continuous physiological monitoring
- automated treatment planning
- adaptive software and machine-learning updates
- clinical workflow copilots
- generative AI and foundation-model-based systems
- patient-facing AI interfaces connected to regulated devices
The regulatory question will increasingly shift from "Does this device use AI?" to a broader set of questions:
That is a much more useful framework than simply counting the number of AI-enabled devices.
11. AI Medical Devices Are Becoming Part of Mainstream Healthcare
The September 2026 FDA update provides a useful snapshot of just how far medical AI has progressed.
More than 1,600 AI-enabled medical devices are now represented in the FDA's public database. Radiology remains by far the largest category. Cardiovascular, neurology, anesthesiology, gastroenterology-urology and hematology are also represented, while the broader list covers additional specialties.
But the most important story is not the number itself.
The important story is that AI is moving from experimental software into regulated medical products that are used for real clinical purposes.
At the same time, regulation is evolving. The FDA is developing approaches for continuously changing AI systems, clinical decision-support software and generative-AI-enabled medical devices.
That means the next stage of healthcare AI will be determined not simply by algorithmic capability, but by clinical evidence, regulatory science, transparency, monitoring and real-world performance.
Frequently Asked Questions
How many AI medical devices has the FDA authorized?
The FDA currently states that it has authorized over 1,600 AI-enabled medical devices for marketing in the United States as of September 2026. A September 2026 analysis of the FDA dataset by Bertalan Meskó identified 1,614 entries.
Are all 1,614 devices FDA approved?
No. The current dataset includes different regulatory pathways. In the September 2026 analysis, 1,553 entries were 510(k), 40 were De Novo and 21 were PMA. Only the PMA figure represents devices approved through the formal Premarket Approval pathway.
Which specialty has the most FDA-authorized AI devices?
Radiology. Meskó's September 2026 analysis identified 1,230 radiology entries, substantially more than any other specialty or category in the highlighted data.
Why are there so many AI medical devices in radiology?
Radiology generates large amounts of structured imaging data, making it particularly suitable for machine-learning applications involving detection, classification, segmentation, reconstruction and quantitative analysis.
Does FDA authorization prove that an AI device is clinically superior?
No. FDA authorization relates to the regulatory requirements applicable to the specific device and pathway. It should not automatically be interpreted as proof of superiority over every competing technology or clinical approach.
Has the FDA approved a generative AI medical device?
The September 2026 analysis by Meskó reported that no device in the analyzed FDA database was identified as using generative AI or being powered by an LLM. However, the FDA is actively developing methods to identify foundation-model functionality and published an August 2026 discussion paper on the regulation of generative-AI-enabled medical devices. The current database should therefore not be interpreted as a complete classification of all generative-AI functionality in regulated medical products.
Can an AI medical device change after FDA authorization?
Potentially. The FDA has established a framework for Predetermined Change Control Plans that can support certain planned AI-enabled device software modifications while maintaining reasonable assurance of safety and effectiveness.
Is the FDA AI-enabled medical device list complete?
No. The FDA explicitly states that the list is not a comprehensive resource of every AI-enabled medical device. It primarily identifies devices using AI-related terms in authorization summaries or classifications.
Sources and Further Reading
- FDA — Artificial Intelligence-Enabled Medical Devices: Official FDA AI-Enabled Medical Devices database and overview
- FDA — 510(k) Premarket Notification. How the FDA evaluates substantial equivalence and 510(k) submissions
- FDA — Premarket Approval (PMA). FDA explanation of the PMA pathway
- FDA — Clinical Decision Support Software. Final FDA guidance issued in 2026
- FDA — Predetermined Change Control Plans for AI-Enabled Device Software. Final FDA guidance on planned AI software changes
- FDA — Generative AI-Enabled Medical Devices. August 2026 discussion paper and request for feedback
- The Medical Futurist — September 2026 analysis. The Medical Futurist's analysis of the FDA AI-device dataset
- Bertalan Meskó, MD, PhD. The Current State Of Over 1600 FDA-Approved, AI-Based Medical Devices. Linkedin
Editorial Note
This article updates and expands the September 2026 analysis of FDA-authorized AI-enabled medical devices. Terminology has been standardized to distinguish FDA clearance, De Novo classification and PMA approval. Numerical figures may change as the FDA updates its database.
This article is for educational and informational purposes and is not medical, legal or regulatory advice.
.png)
.png)
Comments