Brooke & Associates (M. Jason Brooke)

See attachment

Docusign Envelope ID: FC61C361-701E-82EB-8115-C03D106B6CAA

October 19, 2026

BY ELECTRONIC SUBMISSION

Dockets Management Staff (HFA–305)
Food and Drug Administration
5630 Fishers Lane, Rm. 1061
Rockville, MD 20852

RE: Comments Regarding Considerations for the Regulation of Generative AI-Enabled
Medical Devices: Discussion Paper and Request for Feedback (Docket # FDA-2026-N7874)

To Whom It May Concern:
Thank you for the opportunity to submit to the FDA these comments on the recent discussion paper
titled Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion
Paper and Request for Feedback1 (the “Discussion Paper”) on behalf of Brooke & Associates.2

Brooke & Associates is a boutique law and advisory firm that provides legal and consulting
services to innovative medical device, digital health, and digital therapeutics companies across the
globe and at various stages of commercialization. We advise clients of all sizes and clinical,
therapeutic, and technological domains to address issues across a broad spectrum—from
counseling on compliance with legal and regulatory requirements to providing practical guidance
on product-specific market authorization strategies to supporting post-market surveillance
activities. Our team represents a diverse set of stakeholders reflective of the evolving device
industry and includes attorneys, scientists, engineers, clinical and regulatory affairs professionals,
government affairs professionals, and former FDA personnel. Our firm has a particular focus on
hardware and software medical devices incorporating wearable sensors, artificial
intelligence/machine learning, and virtual, augmented, and mixed reality technologies. In this
domain, we have served as an advocate for development of smart regulatory policy for more than
a decade. To that end, we respectfully submit the comments in this letter and the attached Table 1
for consideration by the Agency during its effort to develop policy regarding the regulation of
medical devices that rely on or incorporate generative artificial intelligence (“GenAI”) technology.
U.S. FOOD & DRUG ADMIN., CONSIDERATIONS FOR THE REGULATION OF GENERATIVE AI-ENABLED MEDICAL
DEVICES: DISCUSSION PAPER AND REQUEST FOR FEEDBACK (2026), available at
https://www.fda.gov/media/194242/download [hereinafter GENAI DISCUSSION PAPER].
These comments solely represent the opinions of Brooke & Associates and do not represent the views of any of the
firm’s clients.

-1Docusign Envelope ID: FC61C361-701E-82EB-8115-C03D106B6CAA

GenAI Discussion Paper Comments
Docket # FDA-2026-N-7874

We appreciate the FDA’s effort to propose a baseline policy approach for regulating GenAIenabled devices and providing an opportunity for public comment. However, we have significant
concerns about the conceptual approach presented, specifically with respect to the safety risk
assessment and pre- and post-market monitoring. Fundamentally, our concerns stem from the fact
that the proposed approach is neither novel nor specific to GenAI-enabled medical devices and
contains inherent deficiencies that must be further considered and addressed with a new approach
that is consistent with long-standing principles of the medical device industry as well as takes into
consideration the unique aspects of AI-enabled medical devices, including GenAI-enabled
software functions.

I. THE PROPOSED SAFETY RISK ASSESSMENT
The safety risk assessment framework contemplates a two-dimensional analysis based on 1) the
software output’s influence on clinical decisions and 2) the significance of the safety impact of an
erroneous output. There are several inherent problems with this approach. The proposed risk
framework is inconsistent with well-accepted, industry best practice for risk management for
medical devices. As indicated in the FDA-recognized consensus standard, ISO 14971:2019, risk
is defined as the “combination of the probability of occurrence of harm and the severity of that
harm”, where harm is “injury or damage to the health of people, or damage to property or the
environment” and severity is the “measure of the possible consequences of a [potential source of
harm]”. The probability of occurrence of harm is the combination of the probability that a
hazardous situation would occur as a result of potential source of harm and the probability that the
hazardous situation would lead to harm. This framing of risk is well-established and universally
adopted with the global medical device industry.

Unfortunately, the two-axis risk framework presented for GenAI-enabled software functions in
Figure 1 of the Discussion Paper appears to throw this conventional wisdom out the window.
Specifically, the risk framework replaces the general probability of occurrence of harm parameter
with a crude, qualitative variable that reflects a specific set of clinical situations. This approach
conflates a measure of the likelihood that harm will occur (which, as noted above, depends on both
the likelihood that a hazardous situation will occur and the likelihood that the hazardous situation
will lead to harm) with a measure of inherent risk associated with different clinical situations
resulting from the intended use of the GenAI-enabled software function. More specifically, the
proposed approach assumes a certain probability of harm associated with specific intended uses—
that is, providing information in a non-directive or action-directing manner or taking action in an
HCP-supervised or fully autonomous manner. This activity-based dimension not only incorrectly
accounts for the critical probability of harm element of the risk determination, but it also appears
to duplicatively, albeit erroneously, incorporate the severity of the harm element that is already
contemplated in the “consequences” dimension. For example, a software function that merely
provides information in a non-directive manner about how to manage acute, suicidal ideation in a
patient suffering from severe depression has inherently greater risk than a software function that
autonomously triggers a routine measurement of blood pressure in a patient with mild
hypertension. The proposed risk framework incorrectly suggests that the hypothetical nondirective informational software function has inherently less risky (i.e., has less severe
consequences and is less likely to occur) than the hypothetical fully autonomous, action-taking
software function.

-2Docusign Envelope ID: FC61C361-701E-82EB-8115-C03D106B6CAA

GenAI Discussion Paper Comments
Docket # FDA-2026-N-7874

In addition, the proposed approach is remarkably similar to the SaMD Framework for Risk
Characterization published in 2014 by the International Medical Device Regulators Form
(“IMDRF”), with the exception that the approach proposed in the Discussion Paper offers a
continuous measure as opposed to discrete risk categories. The practical implication of the
continuous measure is that predictably determining where a particular device falls in the
framework and what regulatory oversight applies will be nearly impossible without an exhaustive
list of examples, which is an impractical way to establish regulatory policy. To the extent that the
FDA intends to rely on the IMDRF SaMD Framework for Risk Characterization, the Agency
should review feedback previously provided in Docket No. FDA-2017-D-6569 to avoid
reincarnating the same or similar deficiencies.3

II. THE COMPETENCY-BASED PRE-MARKET EVALUATION
The Discussion Paper indicates that the Agency “is considering a competency-based approach to
premarket evaluation of GenAI-enabled devices that is also inspired, at a high level, by how human
clinicians are evaluated and credentialed, but would be adapted for the technical, practical, and
legal considerations applicable to the regulation of medical devices.” 4 The FDA divides the
competency-based pre-market evaluation approach into 1) non-clinical benchmarking and 2)
clinical confirmation; however, the approach does not appear to deviate from the existing premarket review process, begging the question of what is unique about GenAI-enabled devices that
warrants a different approach and how competency is being integrated.

Fundamentally, the proposed approach involves testing the device using clinical and non-clinical
methodologies, as is the standard today. The “benchmarking elements” described in Appendix A
focus on ways to ensure the device “avoids causing harm”, “possess[] accurate clinical
knowledge”, and maintains “safety and clinical proficiency . . . across foreseeable variation”.5 At
bottom, these “benchmarking elements” simply reflect ways to conduct verification testing.
Likewise, the “clinical confirmation” approaches reflect different ways of conducting validation
testing. Conceptually, the competency-based pre-market evaluation proposal is neither new nor
unique to GenAI-enabled devices. The proposed concept does, however, appear to offer some
greater degree of flexibility as it relates to how to evaluate safety and effectiveness. For example,
the use of shadow deployment would allow for release of investigational aspects of the device in
a real-world scenario for the specific purpose of gathering data without obtaining premarket
authorization. At the same time, the concept begs the question of whether and how other regulatory
requirements apply. Hence, it is important for the FDA to clarify whether and to what extent
regulatory requirements established for investigational device exemption,6 protection of human
subjects, 7 and institutional review boards 8 apply to the clinical confirmation approaches
(especially shadow deployment).
We previously provided comments to IMDRF SaMD Framework for Risk Characterization docket (FDA-2017-D6569-0108) that described significant deficiencies in the proposed approach. Such comments should be incorporated
by reference to these comments given the similarities between the two-dimensional risk assessment frameworks in the
Discussion Paper and the IMDRF SaMD Framework for Risk Characterization guidance.
GENAI DISCUSSION PAPER at 10.
Id. at 24-25.
21 C.F.R. Part 812.
Id. Part 50.
Id. Part 56.

-3Docusign Envelope ID: FC61C361-701E-82EB-8115-C03D106B6CAA

GenAI Discussion Paper Comments
Docket # FDA-2026-N-7874

III. POST-MARKET MONITORING
The lack of substantive discussion of post-market monitoring approaches in the Discussion Paper
makes it difficult to provide feedback to the FDA on a particularly critical aspect of the overall
regulatory framework. Historically, the FDA has struggled to define an innovative, universally
applicable approach to gathering and analyzing data for post-market monitoring of medical
devices. The Agency attempted to develop a novel approach to post-market monitoring for SaMD
products during the Software Precertification (“Pre-Cert”) Pilot Program over the course of five
years from 2017 to 2022.9 Unfortunately, the Agency never proposed a comprehensive, innovative
way to evaluate performance after product deployment. To the extent that the Agency intends to
pursue development of a post-market monitoring program for GenAI-enabled software functions,
prior comments provided as part of the Pre-Cert program should be considered.10

***

Beyond the more general comments described above, we have provided a number of line-item
comments in Table 1 attached.

Thank you in advance for your time and consideration of these comments. If you have any
questions or require additional information, please do not hesitate to contact me via email at the
contact information provided below.

Sincerely yours,

M. Jason Brooke, MSE, JD, CSQE
Attorney & Managing Member
Brooke & Associates
jbrooke@devicecounsel.com
U.S. FOOD & DRUG ADMIN., THE SOFTWARE PRECERTIFICATION (PRE-CERT) PILOT PROGRAM: TAILORED TOTAL
PRODUCT LIFECYCLE APPROACHES AND KEY FINDINGS (2022), available at
https://www.fda.gov/media/161815/download.
We previously provided comments regarding post-market monitoring as part of the Pre-Cert program (Docket No.
FDA-2017-N-4301-0077). Such comments should be incorporated by reference to these comments.

-4Docusign Envelope ID: FC61C361-701E-82EB-8115-C03D106B6CAA

GenAI Discussion Paper Comments
Docket # FDA-2026-N-7874

Table 1: Specific Line-Item Comments on the GenAI Discussion Paper

# Page # Discussion Paper Language Comments
1 Page 1 [T]his paper is not intended to address It is absolutely critical that the FDA contemplate the legal authorities that currently exist and
whether the approaches discussed below whether new legal authorities would be necessary to implement changes to the regulatory
are within FDA’s existing legal authorities framework for GenAI devices. Once the Agency has established its initial thinking on the
or whether new legal authorities would be regulatory framework, the FDA must present the legal basis for its proposed framework and
necessary. get input from the public so that all stakeholders have an opportunity to comment not just on
the regulatory framework but whether the Agency’s legal reasoning is adequately supported.
2 Page 1 Generative artificial intelligence The Agency should clearly and comprehensively describe the unique risks associated with
(GenAI)‑enabled medical devices . . . may GenAI-enabled medical devices so that the public can appropriately consider whether the
introduce unique risks when compared to benefit of the use of these technologies—particularly in light of the FDA’s yet-to-be proposed
traditional software and artificial regulatory framework—outweigh these unique risks. The FDA historically has gone to great
intelligence (AI)-enabled medical devices. lengths to highlight the benefits of innovative technologies without discussing in detail the
associated risks. The Agency would never let a manufacturer of an innovative medical device
place that device on the market without a comprehensive discussion and review of both the
benefits and risks. The FDA should hold itself to the same standard when it comes to
establishing innovative regulatory policy and afford the public an opportunity to fully evaluate
the appropriateness of that policy.
3 Page 1 CDRH’s ultimate goal is to enable a We applaud the FDA’s focus on establishing a least-burdensome approach to regulating
nimble regulatory approach that employs GenAI-enabled medical devices. As the Agency well knows, Section 513(i)(1)(D)(i) of the
least burdensome principles and allows FD&C Act demands that, for example, in the context of a 510(k) premarket review, the FDA
patients timely access to safe and effective “only request information that is necessary to making substantial equivalence determinations”,
medical devices. taking into consideration “the minimum required information”—in other words, “the least
burdensome means of demonstrating substantial equivalence”. The Agency itself defines the
least burdensome principle as requiring “the minimum amount of information necessary to
adequately address a relevant regulatory question or issue through the most efficient manner
at the right time.”11 Accordingly, the goal to enable a nimble regulatory approach is not
simply noble but a legal requirement that the Agency must unquestionably heed.
U.S. FOOD & DRUG ADMIN., THE LEAST BURDENSOME PROVISIONS: CONCEPT AND PRINCIPLES – GUIDANCE FOR INDUSTRY AND FOOD AND DRUG
ADMINISTRATION STAFF (2019), available at https://www.fda.gov/media/73188/download.

-5Docusign Envelope ID: FC61C361-701E-82EB-8115-C03D106B6CAA

GenAI Discussion Paper Comments
Docket # FDA-2026-N-7874

# Page # Discussion Paper Language Comments
4 Page 1 As with all medical devices, CDRH We concur with the FDA that the regulatory policy for GenAI-enabled devices must be riskanticipates applying a risk-based approach based, taking into consideration the product’s intended use and technological characteristics,
to regulating GenAI-enabled devices, consistent with the FD&C Act and the long-standing approach the Agency has taken to
taking into consideration the product’s regulating medical devices.
intended use and technological
characteristics.
5 Pages 7-8 “Action-directing” functions. As noted While we agree that the intended use of the GenAI-enabled software function should dictate
above, certain GenAI-enabled the risk assessment and related regulatory oversight (as is the case for regulating medical
informational functions direct action for a devices today), we disagree that “action-directing” functions involve inherently higher risk
patient or HCP. CDRH is considering than “informational” functions in all cases. As noted above, certain informational functions
treating these “action-directing” functions can have substantially greater risk than some action-directing functions. We share similar
as higher risk functions than GenAI- concerns about the assumptions about inherent risk made in the Discussion Paper regarding
enabled functions that provide simple non- “action-taking” functions, measurement and signal processing functions, patient-facing
directive information. . . . information functions, and HCP-facing functions. Fundamentally, the risk framework must be
structured to facilitate an assessment of risk based on the ISO 14971:2019 and the deviceGeneralist HCP-facing versus specialist
specific intended use.
HCP-facing functions. . . . A function that
helps a generalist HCP approach a question
typically handled by a specialist may
meaningfully extend access to specialty
knowledge, while a function whose safe
use depends on contextualization by
specialist expertise may present elevated
risk when used by HCPs lacking that
expertise to independently evaluate
outputs.
6 Page 15 Shadow deployment. The device operates The use of shadow deployment offers an important method for gathering real-world data to
in a live clinical workflow on real patients, support evaluation of changes to the device. This approach, however, is not unique to GenAIbut its outputs are not shown to clinicians enabled devices and should be applicable to medical devices more generally (especially
or patients and do not affect care. Outputs software-based devices). Importantly, the FDA should clarify how the concept of shadow
are recorded and compared against deployment can practically coexist with other regulatory requirements associated with
decisions actually made and, where deployment of investigational devices.
appropriate, downstream outcomes.

-6Docusign Envelope ID: FC61C361-701E-82EB-8115-C03D106B6CAA

GenAI Discussion Paper Comments
Docket # FDA-2026-N-7874

# Page # Discussion Paper Language Comments
7 Page 15 Clinician adjudication of real cases. It is not clear why the Agency identified clinical adjudication of real cases as a specific
Qualified, independent clinicians review clinical confirmation approach. As it is described in the Discussion Paper, this approach (i.e.,
samples of real patient inputs and device the use of clinical adjudication) appears to apply across the other approaches (e.g.,
outputs against prospectively defined retrospective evaluation, shadow deployment, and prospective clinical study). The FDA
criteria. Depending on the device, this should clarify how this particular approach is unique and different from adjudication methods
might take the form of blinded review, in used in other clinical confirmation approaches.
which a clinician reviews real patient
inputs without seeing the device’s outputs
and independently determines the
appropriate response for comparison
against the device output, or unblinded
review of full interaction records to assess
the quality and appropriateness of outputs.
8 Page 17 CDRH seeks input on what role(s), if any, The FDA should proceed with great caution when considering whether to rely on independent
independent third parties might play in the third parties for premarket evaluation. The decentralized approach to regulation of medical
premarket evaluation approach described devices in the European Union, which relies on independent third parties for pre- and postabove. market regulatory oversight, has demonstrated to be a largely inconsistent and chaotic
approach to regulating. Relying too heavily on third parties in lieu of regulatory functions
otherwise applicable to the FDA would not be in the best interest of patients.

-7-