No, these radiology AI device types cannot skip the 510(k), even when the maker already holds a clearance. FDA denied an exemption petition for four device types on April 1, 2026, and published its final order on September 17, 2026, so each new device needs its own 510(k).

Key takeaways

  • FDA denied the petition by letter on April 1, 2026, and published the final order at 91 FR 58817 on September 17, 2026 [1, 2].
  • The order covers 6 product codes, which match 237 of the 1,614 entries on FDA's AI-enabled device list (15%), including a few PMA and De Novo entries [1, 10].
  • FDA's data shows 241 cleared 510(k)s under these codes, and 150 of them (62%) came from applicants outside the US [9].
  • South Korea has 15 of those clearances, Taiwan 10 and India 7 [9].
  • FDA says makers "may consider" a Predetermined Change Control Plan (PCCP) to reach some of the petition's aims [2].

Can radiology AI skip the 510(k)?

No. FDA's order says manufacturers of these devices "must continue to submit and receive FDA clearance of a 510(k) submission before marketing their devices" [1].

The petition did not ask to remove all review. It asked FDA to exempt later devices from makers that already held a clearance. FDA found that the information did not show a 510(k) was unnecessary to assure safety and effectiveness [1].

The order took effect on September 17, 2026 [1]. It does not change how any device is classified. These device types stay in class II, with special controls [3].

Which devices does the order cover?

The order covers four regulations and six product codes in radiology. All six require a 510(k) [1, 2].

Regulation Device type Short name Product codes
21 CFR 892.2060 Computer-assisted diagnostic software for lesions suspicious of cancer CADx POK
21 CFR 892.2070 Medical image analyzer CADe MYN
21 CFR 892.2080 Computer aided triage and notification software CADt QAS, QFM
21 CFR 892.2090 Computer-assisted detection and diagnosis software CADe/x QBS, QDQ

In plain terms, CADe points the reader to a place on an image, CADx characterizes a finding, and CADt flags a time-sensitive case for faster review [2, 3].

Many AI devices sit outside these codes. For example, image processing software under product code QIH has 305 entries on FDA's AI list, and it was not part of this petition [10]. Our AI device tracker shows the wider picture.

What did the petition ask FDA to do?

The petition asked FDA to exempt a device from 510(k) review when the maker met five conditions. FDA received it on October 22, 2025, from Rubrum Advising on behalf of Harrison.ai [1].

The proposed conditions were [1, 2]:

  • The manufacturer already holds a 510(k).
  • For CADt, that clearance is under 21 CFR 892.2080.
  • For CADe, CADx or CADe/x, the clearance is under any of those three regulations.
  • The manufacturer uses "a robust post-market plan, transparency, and training measures".
  • Existing special controls, quality system, registration and listing rules stay in force.

FDA received more than 45 comments, and a majority opposed the proposal [2]. One concern FDA records is removing oversight of high-risk diagnostic software, "particularly when that software is developed by foreign entities" [2].

Why did FDA say no?

FDA rejected both core conditions, then said it was "not persuaded" on any of its four exemption factors. FDA uses these factors for class II exemptions [2, 4].

Petition's argument FDA's reply [2]
One clearance shows that a maker is proficient A clearance judges one device against one predicate. It does not automatically show the process suits future devices
A clearance in one CAD type should cover other CAD types CADe, CADx and CADe/x work differently. A radiologist may more easily check a box on an image than a malignancy score
A post-market plan can replace review The maker would decide what is proportionate to the risk. FDA says this "would flip the standard" in the statute
Few adverse events are reported A false result alone may not be an adverse event, and the link to later harm may not be clear
CAD has been on the market for 20+ years, so it is well established The 2012 guidances, updated in 2022, cover CADe only, and no vertical consensus or performance standards exist for these types [7]
Users can readily detect changes A reader may not know if an output is right without a biopsy, and may not see performance drift
Changes will not alter the class A new automated feature, indication or user could change the classification, possibly to class III

FDA also noted that CAD and CADt accuracy "can vary greatly depending on the indication" and is "generally much lower than 99 percent" [2]. FDA used that figure to distinguish genetic health risk tests, a type the petition cited as a model.

What does FDA's letter tell you about your own submission?

The letter is useful because it shows what reviewers focus on. Read it before you plan your validation study.

FDA gives one striking fact. Among all 510(k)s cleared under one CAD product code over several years, all were placed on hold. In each case, the reasons included deficiencies in performance testing methodology or results, and in many cases the maker had a prior CAD authorization [2]. FDA does not name the code.

FDA accepts that some practices may carry over, such as good software engineering, cybersecurity and keeping training and validation data independent. But it says these items may not readily transfer from one indication to another [2]:

  • The interdisciplinary expertise needed to understand the workflow, benefits and risks.
  • The data collection protocol for a representative, well-sized dataset.
  • The method used to set the reference standard, or ground truth.

FDA's example is direct. Ground truth methods for dental caries detection may not transfer to a device that detects large vessel occlusion [2]. So plan each indication as its own evidence project.

FDA also notes that a 510(k) holder may not have built the device. It may have bought it or used third parties, so a clearance may not show experience in development [2].

How many AI radiology clearances does this affect?

FDA's data shows 241 cleared 510(k)s under these six codes, with 2025 the highest year so far at 43 [9]. There were 36 in 2026 through mid-October data.

Year Cleared 510(k)s
2018 to 2021 57
2022 36
2023 34
2024 35
2025 43
2026, to date 36

These products are more international than the 510(k) program overall. Since 2018, 62% of clearances under these codes went to applicants outside the US, against about 49% of all 510(k)s [9]. The table below uses the applicant's country.

Applicant country Cleared 510(k)s
United States 91
Israel 39
France 30
South Korea 15
United Kingdom 11
Australia 11
Taiwan 10
Netherlands 10
India 7
Other (8 countries) 17
Japan 0

Review time varies by type. For 114 clearances since January 2024, the median was 146 calendar days from receipt to decision, or 158 without Special 510(k)s [9].

Device type Clearances since January 2024 Median calendar days Median without Special 510(k)s
CADt (892.2080) 49 104 126
CADx (892.2060) 15 138 154
CADe/x (892.2090) 27 159 179
CADe (892.2070) 23 196 196

The petition would have mattered most to repeat filers. Since January 2024, about two thirds of the 114 clearances went to an applicant with an earlier clearance in the same group [9]. The count is 70 to 78, depending on how names are matched. That is our rough measure, and it ignores the petition's other conditions. Reviews like these continue under this order.

What can you do instead of a new 510(k) for every change?

The main option FDA names is a PCCP. FDA says makers "may consider use of a PCCP to achieve some of the aims proposed in the petition" [2]. It encourages a Q-Submission to discuss this [2].

A PCCP is reviewed as part of a marketing submission. It describes planned future changes, so you can make them without a new submission for each one [2, 5]. FDA first issued its guidance on PCCPs for AI devices on December 4, 2024, and reissued it on August 18, 2025 [5]. See our PCCP examples post for what FDA has shown so far.

Change you plan Likely route to test first
Same model, retrained within the specifications and protocol FDA authorized PCCP, if it was authorized in your 510(k) [5]
A change not covered by your PCCP Assess it under FDA's software change guidance [6]
A new finding or body region Assume a new 510(k). A PCCP should generally stay within your indications for use [5]
A new user group, or a move from aiding to automating Talk to FDA first. FDA says this might change the classification [2]

A PCCP has limits. Planned changes must keep the device within its intended use [5]. Some commenters told FDA that PCCPs are not well suited to "meaningful feature expansion", and others said that predicting specific changes does not fit the pace of AI [2]. FDA said it appreciates both views, and it believes PCCPs will help as both sides gain experience [2].

If your roadmap has several indications, each one needs its own evidence plan. Complizen's regulatory experts with direct FDA submission experience can test that plan with you in a gap assessment, before you pay for a reader study.

What should manufacturers outside the US do now?

Plan on one 510(k) for each new device, and build your evidence to FDA's standard from the start. FDA says it does not recognize regulatory authorizations from other countries [12].

  1. Find your product code and confirm which of the four regulations applies [1].
  2. Read the special controls in that regulation. They set required testing and labeling content [3].
  3. Check that your dataset is representative and well sized for each indication. FDA's letter treats datasets and reference standards as indication-specific [2].
  4. Budget the FDA fee: $28,653 for a 510(k) in fiscal year 2027, or $7,163 with small business status [11]. Use our fee calculator for the current numbers.
  5. Decide on a PCCP before you file, because FDA reviews it in the marketing submission [5].
  6. Consider a Pre-Submission if your testing plan or indication is new [2, 8].

Our guide for AI devices from Korea, Japan and Taiwan covers the steps by market. The wider SaMD and AI path lists related reading.

Could FDA change this later?

Yes, the door is not fully closed. The statute lets FDA exempt a class II device on its own initiative or on a petition [1]. FDA says it "intends to continue to consider and develop appropriate approaches" to digital health regulation [2].

FDA also "supports the continued consideration of innovative and least burdensome approaches" [1]. Those words do not promise an exemption, and they are not a change in the rule. For planning, treat the 510(k) as the rule for these device types.

How we counted

We pulled openFDA 510(k) records for product codes POK, MYN, QAS, QFM, QBS and QDQ from the bulk file dated October 30, 2026 [9]. That gave 244 records: 241 substantially equivalent decisions and 3 De Novo grants, which we excluded from the 510(k) counts.

Country means the applicant's address in FDA's data. It may differ from where a company develops its software. Days run from date received to decision date and include hold time.

For repeat filers, we normalized applicant names, then checked for an earlier clearance in the same group. The count includes Special 510(k)s. We treated 892.2080 as one group and 892.2060, 892.2070 and 892.2090 as another, as the petition did. Name changes and acquisitions can affect this count.

For the AI list, we counted rows on FDA's page with these product codes. The page is dated September 4, 2026 [10].

Frequently asked questions

Did FDA exempt AI radiology software from 510(k) review?

No, FDA denied the petition by letter on April 1, 2026, and published a final order on September 17, 2026. Manufacturers of the covered devices must still submit a 510(k) and receive clearance before marketing them. The order is at 91 FR 58817.

Who filed the petition for the 510(k) exemption?

Rubrum Advising, LLC filed it on behalf of Harrison.ai, and FDA received it on October 22, 2025. FDA published a notice on December 29, 2025 and took comments until February 27, 2026. FDA received more than 45 comments, and a majority opposed the proposal.

Which product codes does the FDA order cover?

The order covers POK, MYN, QAS, QFM, QBS and QDQ. These fall under 21 CFR 892.2060, 892.2070, 892.2080 and 892.2090. FDA's AI list has other radiology product codes that were not part of this petition, so check your own code first.

Does a prior 510(k) let me skip review for my next AI model?

Not for a new device under this order. FDA says clearance of one device does not remove the need for clearance of a different device. A change to your cleared device may fit a PCCP or FDA's software change guidance, depending on the change.

Can a PCCP replace a new 510(k)?

A PCCP can cover future modifications that FDA reviewed in your marketing submission, without a new submission for each one. It works only for the changes it describes. FDA says makers may consider a PCCP to reach some of the petition's aims, and encourages early discussion.

Why did FDA treat CADe and CADx differently?

FDA says these devices do different jobs. CADe directs attention to part of an image, while CADx characterizes a finding. FDA notes a radiologist may more easily check a box on an image than a malignancy score, so one clearance does not support the other.

How long does FDA take to clear radiology AI?

The median was 146 calendar days for 114 clearances since January 2024, by our count of openFDA data. The range by type runs from 104 days for CADt to 196 for CADe. These are past results, not a forecast for your device.

Are foreign manufacturers affected more by this order?

The order applies equally to all makers, but more applicants come from outside the US: 150 of 241 clearances, or 62%. FDA's letter also records comments about software from foreign entities. FDA did not make this a stated reason for the denial.

Does an MFDS or CE approval help with these devices?

No, not as an authorization. FDA says it does not recognize regulatory authorizations from other countries, so you still need a 510(k). Some of your test data may help if it meets FDA's expectations, including the population, reference standard and study design for your indication.

How much does the 510(k) cost for an AI device?

The FDA user fee is $28,653 for fiscal year 2027, from October 1, 2026. The small business fee is $7,163. The fee is the same for AI and non-AI devices. Validation data and reader studies add further cost, and that varies by indication.

Could FDA exempt radiology AI in the future?

It could, because the statute lets FDA exempt class II devices on its own initiative or by petition. FDA says it intends to keep developing approaches for digital health. That is not a promise, so plan on the 510(k) for these device types.

Where can I read FDA's full reasons?

FDA's response letter dated April 1, 2026 gives the full reasons in 16 pages. It sits in docket FDA-2025-P-5560 on regulations.gov. The Federal Register order at 91 FR 58817 is much shorter and points readers to that letter for the details.

References

  1. FDA. Medical Devices; Exemption From Premarket Notification: Radiology Computer-Aided Detection and/or Diagnosis Devices and Computer-Aided Triage and Notification Devices, final order. 91 FR 58817, September 17, 2026. https://www.federalregister.gov/documents/2026/09/17/2026-19074/medical-devices-exemption-from-premarket-notification-radiology-computer-aided-detection-andor
  2. FDA, CDRH. Final response letter to Rubrum Advising, LLC, docket FDA-2025-P-5560. April 1, 2026. https://downloads.regulations.gov/FDA-2025-P-5560-0072/attachment_1.pdf
  3. 21 CFR Part 892, including sections 892.2060, 892.2070, 892.2080 and 892.2090 (2025 edition). https://www.govinfo.gov/content/pkg/CFR-2025-title21-vol8/pdf/CFR-2025-title21-vol8-part892.pdf
  4. FDA. Procedures for Class II Device Exemptions from Premarket Notification, guidance. February 1998. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/procedures-class-ii-device-exemptions-premarket-notification-guidance-industry-and-cdrh-staff
  5. FDA. Marketing Submission Recommendations for a Predetermined Change Control Plan for AI-Enabled Device Software Functions, final guidance. August 18, 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/marketing-submission-recommendations-predetermined-change-control-plan-artificial-intelligence
  6. FDA. Deciding When to Submit a 510(k) for a Software Change to an Existing Device, final guidance. October 25, 2017. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/deciding-when-submit-510k-software-change-existing-device
  7. FDA. Computer-Assisted Detection Devices Applied to Radiology Images and Radiology Device Data, final guidance. September 28, 2022. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/computer-assisted-detection-devices-applied-radiology-images-and-radiology-device-data-premarket
  8. FDA. Requests for Feedback and Meetings for Medical Device Submissions: The Q-Submission Program, final guidance. May 29, 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/requests-feedback-and-meetings-medical-device-submissions-q-submission-program
  9. openFDA. Device 510(k) bulk download dated October 30, 2026, pulled November 2, 2026. https://open.fda.gov/apis/device/510k/
  10. FDA. Artificial Intelligence-Enabled Medical Devices. Content current as of September 4, 2026. https://www.fda.gov/medical-devices/artificial-intelligence-enabled-medical-devices/list-artificial-intelligence-enabled-medical-devices
  11. FDA. Medical Device User Fee Rates for Fiscal Year 2027. 91 FR 48134, July 30, 2026. https://www.federalregister.gov/documents/2026/07/30/2026-15335/medical-device-user-fee-rates-for-fiscal-year-2027
  12. FDA. Importing Medical Devices and Radiation-Emitting Electronic Products into the U.S. https://www.fda.gov/medical-devices/importing-and-exporting-medical-devices/importing-medical-devices-and-radiation-emitting-electronic-products-us