Live from FDA’s list, read Nov 16, 2026. Newest decision on the list: Jun 29, 2026. FDA’s list · CSV
When AI software is a medical device
FDA says it does not regulate AI as such: it regulates medical devices, including AI-enabled ones, under the FD&C Act, taking a risk-based approach that considers the device’s intended use and technological characteristics. Software is a device when it is intended for the diagnosis, cure, mitigation, treatment or prevention of disease, or to affect the structure or function of the body, except for software functions that section 520(o) of the FD&C Act excludes. FDA also reviews modifications to AI-enabled device software functions that could significantly affect safety or effectiveness.
The three pathways, and how many AI devices took each
AI-enabled devices use the same premarket pathways as other devices. Counts are from FDA’s AI-enabled device list, by submission number.
| Pathway | When it applies | On FDA’s AI list |
|---|---|---|
| 510(k) K… | The device is substantially equivalent to a legally marketed predicate: the same intended use, and the same technological characteristics or differences that do not raise different questions of safety and effectiveness. | 1,553 (96%) |
| De Novo DEN… | A novel device with no legally marketed predicate, for which general controls, or general and special controls, give reasonable assurance of safety and effectiveness. Granted De Novo devices are classified into Class I or II and can be predicates for later 510(k)s. | 40 (2%) |
| PMA P… | Premarket approval, FDA’s most stringent review, for Class III devices: those that support or sustain life, are of substantial importance in preventing impairment of health, or present a potential unreasonable risk. | 21 (1%) |
Not sure which fits? 510(k) vs De Novo walks through it, and a Pre-Submission lets you ask FDA before you file.
Where AI devices cluster
76% of the devices on FDA’s list are reviewed by the radiology panel. These product codes carry many of them; each page lists recent clearances and what FDA expects.
| Code | Device type | Regulation | On FDA’s AI list | All 510(k)s (openFDA) |
|---|---|---|---|---|
| QAS | Radiological triage software | 21 CFR 892.2080 | 86 | 90 |
| LLZ | Radiological image processing | 21 CFR 892.2050 | 161 | 2,297 |
| QIH | AI image processing | 21 CFR 892.2050 | 305 | 338 |
| QFM | AI lesion prioritization | 21 CFR 892.2080 | 40 | 41 |
| QBS | AI fracture detection | 21 CFR 892.2090 | 10 | 12 |
| POK | AI lesion characterization | 21 CFR 892.2060 | 22 | 24 |
AI list counts by primary product code, live from FDA’s list, read Nov 16, 2026. 510(k) totals from openFDA, as of Nov 16, 2026; they include clearances of devices without AI. Most common codes on the list: QIH (305), LLZ (161), IYN (124), JAK (103), LNH (100), QAS (86).
Special controls for AI radiology software
Three Class II regulations set out what submissions for AI image-analysis software must include. Each also requires a detailed algorithm description, software documentation and labeling that describes the test data and the device’s limitations.
| Regulation | Device type | Performance testing the special controls require |
|---|---|---|
| 21 CFR 892.2080 | Triage and notification (QAS, QFM) | Pre-specified testing that shows effective triage, such as improved time to review; standalone performance; results across important cohorts with confidence intervals |
| 21 CFR 892.2090 | Detection and diagnosis (QBS and others) | Standalone performance with diagnostic accuracy measures, localization accuracy when the device marks a location, and testing that shows improved assisted-read performance, such as a reader study |
| 21 CFR 892.2060 | Characterizing lesions suspicious for cancer (POK) | Testing that shows improved reader performance in the intended use population, standalone performance, and enough cases per important cohort |
Image processing software under 21 CFR 892.2050, which includes QIH and LLZ, is Class II with special controls and names DICOM, JPEG and the SMPTE test pattern as voluntary standards, without device-specific testing in the regulation.
Predetermined change control plans
Section 3308 of the Food and Drug Omnibus Reform Act of 2022 (FDORA), enacted December 29, 2022, added section 515C, “Predetermined Change Control Plans for Devices”, to the FD&C Act. FDA’s final guidance for AI-enabled device software functions, issued December 4, 2024 and updated August 18, 2025, applies to devices reviewed through the 510(k), De Novo and PMA pathways. FDA reviews the PCCP as part of the marketing submission.
- Description of Modifications: the planned changes to the device.
- Modification Protocol: how each change will be developed, validated and implemented.
- Impact Assessment: the benefits and risks of the planned changes, and their mitigations.
FDA’s 510(k) database can filter for clearances with an authorized predetermined change control plan: search it. FDA, Health Canada and the MHRA also published five guiding principles for PCCPs: focused and bounded, risk-based, evidence-based, transparent, and a total product lifecycle perspective.
Current guidance and principles, with dates
| Document | Status | Date |
|---|---|---|
| Marketing Submission Recommendations for a Predetermined Change Control Plan for AI-Enabled Device Software Functions | Final | Aug 18, 2025 (first issued Dec 4, 2024) |
| Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations | Draft, not for implementation | Jan 7, 2025 |
| Good Machine Learning Practice for Medical Device Development: Guiding Principles | Guiding principles | Oct 2021 (FDA, Health Canada, MHRA); IMDRF final document Jan 2025 |
| Predetermined Change Control Plans for Machine Learning-Enabled Medical Devices: Guiding Principles | Guiding principles | FDA page current as of Aug 18, 2025 |
| Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles | Guiding principles | FDA page current as of Jun 13, 2024 |
| Requests for Feedback and Meetings for Medical Device Submissions: The Q-Submission Program | Final | May 2025 |
The January 2025 draft lifecycle guidance recommends what an AI-enabled device’s marketing submission should contain, section by section: device description, user interface and labeling, risk assessment, data management, model description and development, validation, including performance validation, device performance monitoring, cybersecurity, public submission summary. On validation, it asks whether test data are independent of the training data, for example sampled from different clinical sites, and kept away from the model developers.
Generative AI and large language models
FDA says it will explore methods to identify and tag devices on its AI list that incorporate foundation models, from large language models to multimodal architectures, and encourages sponsors to describe LLM-based functionality in their public summaries. FDA’s Digital Health Advisory Committee met on total product lifecycle considerations for generative AI-enabled devices on November 20–21, 2024, and on generative AI-enabled digital mental health devices on November 6, 2025.
Ask about an AI-enabled device
Bearnier AI answers from FDA’s records, with every precise fact cited to its page.
Questions about FDA and AI-enabled devices
Does FDA regulate AI?
FDA says it does not regulate AI as such; it regulates medical devices, including AI-enabled ones. Software is a device when it is intended for diagnosis, cure, mitigation, treatment or prevention of disease, or to affect the structure or function of the body, except for software functions excluded by section 520(o) of the FD&C Act.
How many AI-enabled medical devices has FDA authorized?
FDA says it has authorized over 1,600 AI-enabled medical devices as of September 2026. Its public list held 1,614 entries when we read it on Nov 16, 2026, the newest decided Jun 29, 2026. FDA notes the list is not comprehensive and is updated periodically.
Can an AI-enabled device be cleared through a 510(k)?
Yes, and most are: 1,553 of the 1,614 devices on FDA’s list have 510(k) numbers, against 40 De Novo grants and 21 PMA approvals.
What is a predetermined change control plan (PCCP)?
A plan, reviewed by FDA as part of a marketing submission, that describes the planned modifications to an AI-enabled device, the methods to develop, validate and implement them, and an assessment of their impact. Section 515C of the FD&C Act, added by FDORA in 2022, provides for PCCPs, and FDA’s final guidance applies to 510(k), De Novo and PMA submissions.
Is FDA’s AI device list complete?
No. FDA says the list is not a comprehensive resource: devices are identified mainly by AI-related terms in their summaries or classification, and devices without a published decision summary are added in later updates.
Does FDA’s AI guidance apply to large language models?
FDA’s guidance covers AI-enabled device software functions generally. On its AI device list, FDA says it will explore ways to identify and tag devices that incorporate foundation models, from large language models to multimodal architectures, and its Digital Health Advisory Committee met on generative AI-enabled devices in November 2024 and November 2025.
Sources
- Artificial Intelligence-Enabled Medical Devices
- List of Artificial Intelligence-Enabled Medical Devices
- List of AI-Enabled Medical Devices (CSV download)
- Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions
- Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations
- Good Machine Learning Practice for Medical Device Development: Guiding Principles
- Predetermined Change Control Plans for Machine Learning-Enabled Medical Devices: Guiding Principles
- Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles
- Premarket Notification 510(k)
- De Novo Classification Request
- Premarket Approval (PMA)
- 21 CFR 892.2050: Medical image management and processing system
- 21 CFR 892.2060: Radiological computer-assisted diagnostic software for lesions suspicious of cancer
- 21 CFR 892.2080: Radiological computer aided triage and notification software
- 21 CFR 892.2090: Radiological computer-assisted detection and diagnosis software
- 510(k) Premarket Notification database
- openFDA device 510(k) API
- Requests for Feedback and Meetings for Medical Device Submissions: The Q-Submission Program