What’s behind thisMethodologyPrepared byRecovryAI
Docket FDA-2026-N-7874Published November 10, 2026

Public response to the FDA on generative AI medical devices

Filers agreed with the FDA’s core proposals and asked for changes. More than half asked for more oversight than the FDA proposed.

530comments
26FDA questions
10takeaways

This research analyzes every public comment on the FDA’s discussion paper on generative AI-enabled medical devices: 530 comments from 509 filers. AI models read and sorted every comment. A reviewer at RecovryAI checked the results, and the published figures and quotes, against the filings. RecovryAI, a company developing FDA-regulated AI for post-operative recovery, prepared this research to bring greater understanding, visibility and discussion to the regulation of AI in health care, and also filed a comment on this docket.

OverviewKey takeaway 1 of 10

The FDA’s largest AI consultation to date

All 509 filers, and comment counts on 11 earlier FDA consultations on AI

What’s behind this

530 commentsGenerative AI medical devices, 2026

  • 239AI in early-phase clinical trials, 2026

  • 155AI/ML device software framework, 2019

  • 124AI in drug regulatory decisions, 2025

More than double the response to any earlier FDA consultation on AI.*

509 filers sent 530 comments, about 1.7 million words, in 63 days. The next largest, a 2026 request on AI in clinical trials, drew 239 comments.

* Largest by comments posted to a single docket on regulations.gov, checked November 8, 2026. The 530 leaves out 10 exact repeats.

What’s behind this

OverviewKey takeaway 2 of 10

Startups were half of the industry response, and several well-known names did not file

All 509 filers by kind, and a name search of every filer and submitter

What’s behind this

509 filers

  • Startups23%
  • Other industry22%
  • Clinicians24%
  • Researchers8%
  • Patients2%
  • Other21%

Filed

Did not file

AI and data

  • Google
  • Microsoft
  • Amazon
  • Anthropic
  • Epic
  • Oracle Health
  • OpenAI
  • Meta
  • xAI
  • Apple
  • NVIDIA

Medical devices

  • Johnson & Johnson
  • GE HealthCare
  • Siemens Healthineers
  • Baxter
  • Philips
  • Abbott
  • Stryker
  • Boston Scientific

Hospitals and telehealth

  • Mayo Clinic
  • Johns Hopkins Medicine
  • Duke Health
  • Cleveland Clinic
  • Teladoc Health

Health insurers

  • Kaiser Permanente
  • UnitedHealth Group
  • Elevance Health
  • CVS Health (Aetna)
  • Cigna
  • Humana
  • Centene
  • Blue Cross Blue Shield
  • AHIP (their trade group)

45% came from industry, and half of those were startups.

Clinicians were 24% of filers and researchers 8%. Patients and patient groups were 2%, 9 filers.

Of all 509 filers; groups from RecovryAI’s coding of each filer. Other: 64 filers who did not say who they were, plus policy groups, standards bodies and members of the public.

What’s behind this

Google, Microsoft and Anthropic filed. OpenAI and most large insurers did not.

9% of all 509 filers named OpenAI or one of its AI products (48 filers), more than named any other AI company. Of eight large US health insurers, only Kaiser Permanente filed.

No filing found in a search of every filer and submitter name, including brands and subsidiaries. Named: a comment mentions the company or one of its AI products.

What’s behind this

OverviewKey takeaway 3 of 10

More than half asked the FDA for more oversight than it proposed

Each filer’s overall position on the FDA’s proposals, by kind of filer

What’s behind this
More oversightLess oversightSmall or no changes
  • Trade groups13 filers15%77%8%
  • Large companies40 filers15%40%42%
  • Startups115 filers41%17%42%
  • All filers479 filers54%14%32%
  • Consultants51 filers69%4%27%
  • Researchers38 filers61%3%37%
  • Clinicians114 filers73%7%18%

54% asked for more oversight than the FDA proposed.*

32% asked for small changes or none, and 14% asked for less.

* Of the 479 filers whose filing took an overall position on the FDA’s proposals; 30 took none. Positions are RecovryAI’s coding.

What’s behind this

73% of clinicians asked for more oversight; 77% of trade groups asked for less.

Trade groups are a small base, 13 filers. Large companies split, 40% for less and 42% for small changes or none. Startups asked for more over less, 41% to 17%.

Out of the filers of each kind whose filing took an overall position: 114 clinicians, 13 trade groups, 40 large companies, 115 startups. Kinds from RecovryAI’s coding.

See all 26 questions

What’s behind this

Deep divesKey takeaway 4 of 10

A disclaimer does not make AI advice safer, filers told the FDA

Questions 2 and 3

What’s behind this

4A disclaimer doesn’t make AI advice safer

96% said a disclaimer doesn’t make AI advice safer.*

That is of a small base, the 75 filers who took a side. One said it does, and two said it partly does.

* Of the 75 filers who took a side on whether a disclaimer lowers the risk of advice that tells a patient what to do. Another 124 who answered Question 2, on how an answer’s wording changes its risk, didn’t take it up.

What’s behind this

30% asked for a way to reach a person or emergency care.

It was the safeguard named most, by 37% of clinicians and 29% of industry filers.

Of the 202 filers who answered Question 3, on safeguards for AI that gives clinical information to patients; counted only where an answer says so.

See Questions 2 and 3

What’s behind this

Deep divesKey takeaway 5 of 10

The FDA’s core proposals won broad support, but most filers wanted changes

The seven core proposals: Questions 1, 7, 9, 11, 16, 20 and 25

What’s behind this

1Filers generally agreed with the FDA’s core proposals, but asked for changes

The 767 answers that gave a yes or no

  • 85%yes, with changes
  • 13%yes, as written
  • 2%no or split

875 more answers gave advice without a yes or no.

The FDA’s seven core proposals

  1. Q1Should risk be rated by what the device does and how much harm a wrong answer could cause?Yes, with changes · 93%Weigh whether a mistake can be undone
  2. Q7Should AI devices be tested on benchmarks first, then confirmed in clinical use?Yes, with changes · 69%Test the device exactly as shipped
  3. Q9Can a set of benchmark tests give enough evidence of safety and effectiveness?Yes, with changes · 95%Use outside standards and specialty societies
  4. Q11Can some devices skip a new study on patients before launch?Yes, with changes · 80%Studies with real patients for the riskiest uses
  5. Q16Should independent third parties help test devices?Yes, with changes · 88%Set rules for who counts as independent
  6. Q20Can AI agents help monitor devices after launch?Yes, with changes · 96%Test the monitoring AI’s own accuracy
  7. Q25Should AI model makers keep a confidential file at the FDA that device makers can cite?Yes, with changes · 86%Set what a model file must contain

Of the 301 filers who gave a yes or no: 87% yes, with changes · 7% as written · 6% no or split*

87% said yes to the FDA’s core proposals, with changes.*

They backed each of the seven proposals they answered but wanted at least one changed. 7% said yes as written, and 6% said no or split.

* Of the 301 filers who gave a yes or no to at least one of the seven; a split is yes for some devices and no for others. Stamps: share of that question’s yes-or-no answers, one per filer.

What’s behind this

Filers said yes, then marked it up.

Each proposal drew its own changes, most often adding safeguards.

In pen: the change asked for most on each proposal.

See Questions 1, 7, 9, 11, 16, 20 and 25

What’s behind this

Deep divesKey takeaway 6 of 10

Closer monitoring in place of some premarket testing got a conditional yes

Question 18

What’s behind this
What the FDA proposed, in its words
“CDRH is considering whether it may be appropriate to accept greater premarket uncertainty regarding a GenAI-enabled device’s benefit-risk profile through greater reliance on postmarket monitoring.”
Yes, with conditions · 91%*

Only if…

  • A monitoring plan and alert levels set premarket33%
  • Proof the monitoring catches problems in time33%
  • A tested way to pause or roll back the device19%

But not for… 49%

  • Acts on its own, with no one checking
  • Makes mistakes that can’t be undone
  • Triage or emergency decisions
  • The monitoring the FDA listed
  • Periodic device benchmarking
  • Periodic sample-based clinician review
  • Performance degradation monitoring

91% would accept closer monitoring in place of some premarket testing, with conditions.*

1% said yes outright and 7% said no.

* Of the 205 filers who gave a yes or no on Question 18. Another 38 of the 243 who answered gave advice without one.

What’s behind this

33% asked for a monitoring plan set premarket, and 33% for proof it works.

19% asked for a tested way to pause or roll back the device.

Of the 243 filers who answered Question 18. A filer could name several; each counted only where an answer says so.

What’s behind this

49% said the riskiest uses must still be tested premarket.

65% of clinicians said so, and 46% of industry filers. Named most often: uses that act with no one checking, or make mistakes that can’t be undone.

120 of the 243 filers who answered Question 18, among them 34 of 52 clinicians and 60 of 130 industry filers; counted only where an answer says so. The uses are ranked by a keyword reading, so they carry no counts.

See Question 18

What’s behind this

Deep divesKey takeaway 7 of 10

The device maker stays responsible, a majority said, even for AI it did not build

Questions 21 and 24

What’s behind this

Device maker

A device built on another company’s model

67%Responsible
after launch

Model maker

AI foundation model

Tested versionNew version

Asked for advance notice of model changes53%

  1. Re-test a new version before patients get it58%
  2. Track the version and catch any change54%
  3. Lock to the tested version41%
  4. Keep a way to roll back32%

Clinician and patient

Device output

As testedMay change

67% said the device maker stays responsible after launch.

That held even where hospitals help monitor. 11% said whoever controls a change is responsible, and 22% didn’t say.

Of the 251 filers who answered Question 21, on roles after launch.

What’s behind this

The FDA asked about devices built on another company’s AI model.

From its paper: “Devices built on third-party foundation models raise particular change-control considerations because changes to the underlying model may be initiated by the foundation model developer rather than the device manufacturer.”

58% asked the device maker to re-test a new model version before patients get it.

53% asked for advance notice of model changes. Of those 138 filers, 129 also wanted the device maker’s own checks.

Of the 260 filers who answered Question 24, counted only where an answer says so.

See Questions 21 and 24

What’s behind this

Deep divesKey takeaway 8 of 10

How much an AI may do on its own should depend on the risk, filers said

Questions 1 and 26, and the FDA’s two-axis risk chart

What’s behind this

8AI autonomy scales with risk

59% kept the FDA’s two‑axis risk framework as the base.

Its two axes: what the AI does, and how bad a mistake could be.

Of the 316 filers who answered Question 1. The chart is Figure 1 of the FDA’s discussion paper.

What’s behind this

71% asked the FDA to also weigh whether a mistake can be caught or undone.

Whether a mistake can be undone led at 49%; safeguards further down the line came next at 42%.

Of the 316 filers who answered Question 1. Many named several factors, so the rows add up to more than 71%.

What’s behind this

38% asked for a person’s sign‑off before an AI agent acts.

Records of what the agent did, testing of the whole chain and limits on what it can do each drew about a third.

Of the 232 filers who answered Question 26, on AI agents; 137 named two or more.

See Questions 1 and 26

What’s behind this

Deep divesKey takeaway 9 of 10

No single standard emerged for measuring AI devices

Question 14

What’s behind this

9No single standard emerged for measuring AI devices

An AI device should be measured against

  1. 36%different standards for different risks or devices
  2. 26%the clinicians who will use it
  3. 19%usual care, without the AI

The FDA’s own examples

  • 9%the standard of care
  • 9%specialists or expert panels
  • 2%a typical (“median”) clinician
  1. 59%test the clinician and the AI together
  2. 24%also wanted the AI’s own results reported

No standard for measuring an AI device drew more than 36%.*

The top answer was different standards for different risks or devices. The three options the FDA named each drew under 10%.

* Of the 129 filers who named a default standard; the other 112 of the 241 who answered Question 14 named none. Separately, 49 asked for specialists to judge specialist tasks.

What’s behind this

59% asked to test the clinician and the AI together.

The FDA also asked whether to test the AI alone or with the clinician using it. 24% also wanted the AI’s own results reported.

Of the 241 filers who answered Question 14; each counted only where an answer says so.

See Question 14

What’s behind this

Deep divesKey takeaway 10 of 10

Privacy, security and liability led the issues the FDA’s paper does not name

Every filing read for content

What’s behind this

FDA discussion paper

Considerations for the Regulation of Generative AI-Enabled Medical Devices

  1. Assessment of risk6 questions
  2. Premarket evaluation11 questions
  3. Postmarket monitoring7 questions
  4. Foundation models and agents2 questions

46% of filers raised at least one of these

None is named in the FDA’s paper

  • 15%Privacy and the use of patient data
  • 13%Cybersecurity and data security
  • 11%Who is liable when harm occurs
  • 10%The cost to small companies
  • 10%Staff workload
  • 2%Who pays for the device

46% raised an issue the FDA’s paper doesn’t name.

Patient data privacy led at 15%, then cybersecurity at 13% and liability for harm at 11%.

230 of the 504 filers whose filings could be read for content; five filings were set aside, four not readable for content and one restricted. None of these words appears in the paper: privacy, security, liability, cost, workload, reimbursement.

See all 26 questions

What’s behind this

How this was done

Published November 10, 2026, version 1.0. Covers every comment the FDA posted through November 3, 2026.

  1. Every filingAll 530 filings from 509 filers, read in full, about 1.7 million words.
  2. Sorted the same wayAI models read and sorted every comment against the same written categories. The categories were fixed before coding and applied the same way whether a filer asked for more oversight or less. Each decision had to quote the filing word for word.
  3. Checked by a reviewerA reviewer at RecovryAI checked the results, and the published figures and quotes, against the filings.
  4. Checked between two modelsAnthropic’s Claude coded every filing, and an OpenAI model coded about a fifth independently. On the 65 filings measured, the two gave the same verdict on the FDA’s proposals 96% of the time, and 81% where either found a yes or no.
  5. Out of those who answeredNot every filer answered every question, so each figure is out of those who answered it.
  6. RecovryAI’s own commentRecovryAI also filed a comment on this docket. It is counted in the figures like any other filing, but its words are never quoted or used as an example, and without it no headline figure moves by more than half a point.

Read the full methodologyResults by question

Found an error? Write to support@recovry.ai. Corrections will be listed here with the date they were made.

About RecovryAI

RecovryAI is developing FDA-regulated AI for post-operative recovery. This research is intended to bring greater understanding, visibility and discussion to the regulation of AI in health care.

recovry.ai