Combination Products Coalition (Bonnie Odom)

Please find attached comments from the Combination Products Coalition.

1227 25TH ST. NW, SUITE 700
WASHINGTON, DC 20037

VIA ELECTRONIC SUBMISSION

October 19, 2026

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

Re: Docket No. FDA-2026-N-7874 Considerations for the Regulation of Generative AI-Enabled Medical
Devices: Discussion Paper and Request for Feedback

Dear Sir or Madam:

The Combination Products Coalition (“CPC”)1 welcomes the opportunity to provide comments on the Agency’s
Discussion Paper and Request for Feedback on “Considerations for the Regulation of Generative AI-Enabled
Medical Devices” dated August 18, 2026 (the “Discussion Paper”).

CPC greatly appreciates the Agency’s efforts to provide the Discussion Paper which addresses a timely and
important topic for both the medical technologies sector at large as well as our coalition, which is comprised of
member companies developing a range of products including combination products with embedded firmware
and software applications which may be labeled for use with drug or biological products. Relevant examples of
Generative AI-Enabled Medical Devices that are of interest to CPC member companies include, but are not
limited to:

 Chatbots which assist patients in managing symptoms and side effects of medicines to understand
available mitigations and when contact with a healthcare provider is advisable;

 Remote patient monitoring solutions with embedded models used within medicine studies to synthesize
multimodal sources such as physiologic data and symptom surveys and apply smart thresholds or
screening criteria to indicate potential safety signals or sequelae of clinical interest; and

 Disease risk models which intake patient health information and interpret established diagnostic
guidelines to suggest if patient may be at risk of developing a disease or extent of existing diagnosis.
CPC is a group of leading drug, biological product, and medical device manufacturers with substantial experience and interest in
combination product issues. One of our top priorities is to work collaboratively with the Agency on issues affecting combination
products and digital products to advance our common mission: providing the best possible health care to patients. Our diverse, crossindustry membership permits CPC to bring a special, broad, and unique perspective to these issues.

COMBINATION PRODUCTS COALITION PAGE 1 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

CPC’s comments on the Discussion Paper are captured below. The comments are organized by “Discussion
Question” number as presented in the Discussion Paper. Where a comment is not aligned with a specific
“Discussion Question,” it is categorized as a “General Comment.”

Overall, CPC supports a risk-proportionate, lifecycle-based approach to oversight of generative artificial
intelligence (“GenAI”)-enabled medical devices. FDA policy should be coordinated across the relevant Centers
and the Office of Combination Products, distinguish lower-risk functions that may be appropriate for enforcement
discretion or statutory exclusion, and provide flexible evidentiary pathways—including benchmarking, clinical
simulation, and, where appropriate, synthetic data. These pathways should be aligned with the device’s intended
use, level of patient risk, and stage in the product lifecycle.

General Comments

1. The Importance of Cross-Center Collaboration Regarding Regulation of GenAI-Enabled Devices.

CPC recognizes that the Center for Devices and Radiological Health (“CDRH”) is the Agency’s medical
product center responsible for the regulation of medical devices. However, it is important to note that
GenAI-enabled devices may represent device constituent parts of combination products either as a drugdevice or biologic-device combination product. GenAI-enabled medical devices may also be used as tools
in investigational studies for drug or biologics and may be additionally categorized as remote data
acquisition software or digital endpoints.

While we generally applaud CDRH for releasing the Discussion Paper on GenAI-enabled medical devices
we strenuously encourage CDRH to solicit input from and actively collaborate with other medical product
centers and the Office of Combination Products in any further developments of policy, guidance, or
regulation in the GenAI space.

CPC also encourages the Agency to maintain consistency between future GenAI-enabled medical device
policies and existing and evolving cross-Center guidance applicable to digital health technologies, such
as the Agency’s “Digital Health Technologies for Remote Data Acquisition in Clinical Investigations”
guidance.2 Such alignment is particularly important where a GenAI-enabled technology may be used in a
drug or biologic clinical investigation while also potentially being subject to medical device requirements.

CPC requests that the Agency clarify how the proposed GenAI framework will apply when a GenAI-enabled
device function is a constituent part of, or is cross-labeled for use with, a drug or biological product. Given
that the performance of a GenAI-enabled device function may be influenced by interactions with product
constituents, delivery systems, users, clinical workflows, and external data sources, additional clarity
regarding how these factors may be considered in the combination product context, including as part of
benefit-risk assessment, validation, labeling, change control, and postmarket monitoring of the
combination product as a whole, would be valuable. Additional discussion or examples in the combination
product context would help stakeholders understand how the proposed framework may be applied to
these integrated products and support consistent implementation across FDA Centers.
Available at https://www.fda.gov/media/155022/download.

COMBINATION PRODUCTS COALITION PAGE 2 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

2. Recognizing that Certain GenAI-Enabled Medical Devices May Qualify for Exclusion from the Definition of
a Medical Device under 21st Century Cures Act Provisions Regarding Clinical Decision Support Software.

Within Section III of the Discussion Paper, Definitions and Context, the Agency references several
precedent guidance documents and policies related to the regulation of digital health technologies and
software as a medical device. CPC considers that a significant percentage of GenAI-medical device use
cases will likely be considered as “Clinical Decision Support Software” and further that such software may
be exempt from FDA medical device regulation if all four exclusionary criteria outlined in Section
520(o)(1)(E) of the Federal Food, Drug, and Cosmetic Act are met. In future publications, the Agency
should formally recognize that some GenAI systems may be considered to be non-device clinical decision
support software. Additional guidance should be provided on how the Agency intends to interpret the
statutory exemption requirement “enabling such health care professional to independently review the basis
for such recommendations that such software presents so that it is not the intent that such health care
professional rely primarily on any of such recommendations to make a clinical diagnosis or treatment
decision regarding an individual patient” in the context of GenAI-enabled medical devices.

CPC would also welcome additional clarity regarding when agentic AI-enabled functions are expected to
be considered device functions versus non-device functions. Given that the Discussion Paper
acknowledges that many healthcare uses of agentic AI may fall outside the Agency's device regulatory
oversight, additional discussion or examples would help sponsors better understand the scope of
applicability of the proposed framework.

3. Consideration for the Need for Formal Rulemaking to Realize Changes in Regulations to Account for GenAI
and other AI Technologies.

CPC understands that the Discussion Paper was not intended to address “whether the approaches
discussed…are within the Agency’s existing legal authorities or whether new legal authorities would be
necessary.” However, given the expansive nature of some of the Agency’s discussion topics which would
seem to impact areas of existing regulation such as classification procedures under 21 CFR Part 860,
premarket submissions under 21 CFR 807 and 814, as well as Quality Management System Regulation
under 21 CFR 820, CPC encourages the Agency to provide greater transparency regarding the legal and
regulatory mechanisms that would be necessary to implement any future policy changes arising from this
discussion.

CPC is generally supportive of the changes the Agency may be considering and are covered within the
Discussion Paper for GenAI-enabled medical devices; however, we suggest that the Agency also
addresses the potential need for and probability of success of rulemaking in realizing these changes. This
transparency would greatly assist the industry and public at large in understanding the timescale and level
of effort required to enact significant change in the regulation of GenAI-enabled medical devices.

COMBINATION PRODUCTS COALITION PAGE 3 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

4. Harmonization with Other Digital Health Regulatory Frameworks.

CPC further encourages the Agency to consider existing internationally recognized risk-based approaches
for GenAI-enabled devices and to promote regulatory alignment where practical, particularly the
International Medical Device Regulators Forum (IMDRF) Software as a Medical Device (SaMD) risk
categorization framework, or to provide a clear mapping between the proposed framework and existing
Agency and international risk constructs. While the proposed device activity and consequence axes appear
conceptually similar to existing approaches used to characterize the significance of information provided
by software and the potential impact of incorrect outputs, introducing a new framework without an explicit
relationship to established paradigms may create ambiguity, duplicative risk assessments, and
unnecessary regulatory complexity. Clear alignment would promote international harmonization, facilitate
regulatory convergence, and enable manufacturers to leverage established risk-management processes
and documentation throughout the product lifecycle.

In addition to alignment across FDA Centers and internationally recognized risk-based frameworks, it will
be important that future GenAI-enabled medical device policies are developed in a manner that is
consistent with other existing U.S. regulatory, policy, and healthcare frameworks that may be applicable
to these technologies such as that proposed in FDA’s draft guidance “Considerations for the Use of
Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products.”3

5. Alignment with Existing Cybersecurity Frameworks.

CPC recognizes that agentic AI-enabled devices may introduce additional cybersecurity considerations
relating to external tool use, prompt injection, model manipulation, and other autonomous behaviors. The
Agency should provide additional clarity regarding how existing cybersecurity principles and riskmanagement approaches apply to these emerging technologies. CPC further encourages the Agency to
align any future expectations with existing FDA medical device cybersecurity guidance, where appropriate,
to promote consistency and avoid duplicative requirements.

Comments Regarding Discussion Paper Section IV: Considerations for the Assessment of Risk
for GenAI-Enabled Devices

1. Does the two-axis risk framework, organized around device activity and the consequence of relying on an
incorrect output, appropriately capture the dimensions most relevant to the risk of a GenAI-enabled software
function? If there are additional dimensions—such as the reversibility of a resulting action, the availability of
downstream safeguards, the time pressure of the deployment setting, or the traceability of the output (i.e., to
primary source materials)—that should be represented in a risk framework, please describe and provide
examples of how they should be represented.

CPC supports the Agency’s general approach in creating an expanded risk-based framework which may be used
to apply regulatory oversight to GenAI-enabled software functions. CPC believes that additional resolution
provided by a multi-axis framework can assist the Agency in a more nuanced approach to regulation and would
Available at https://www.fda.gov/media/184830/download.

COMBINATION PRODUCTS COALITION PAGE 4 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

ideally allow for a continuum of oversight as opposed to the current three level approach to classification (e.g.
Class I, II, III). CPC suggests that it would be appropriate for GenAI-enabled software functions that fall on the
lower risk end of the risk framework be subject to enforcement discretion conditions with increasing regulatory
scrutiny applied as the risk increases.

While CPC supports retaining the proposed two-axis framework, the Agency should consider the extent to which
an incorrect output may propagate through subsequent decisions, workflows, or systems as an additional
consideration when assessing overall risk and determining appropriate oversight.

2. CDRH seeks input on how to account for the spectrum of GenAI-enabled informational functions that vary in
the degree to which they direct a user to a particular action (i.e., between “non-directive” and “action-directing”).
What characteristics of an output—such as its wording, specificity, personalization, or context—could be
considered as modifiers of the risk associated with an informational function, after accounting for the device’s
overall functionality and intended use? What additional information would provide manufacturers with
sufficient clarity and predictability around risk assessment for informational functions while recognizing that
directiveness may exist along a continuum rather than as a binary distinction?

CPC agrees that "directiveness" exists on a continuum rather than as a binary concept and should be assessed
based on the totality of the output and deployment context rather than individual words or phrases. CPC suggests
the Agency should allow directiveness to be characterized through a combination of: (i) the specificity of the
output (i.e., a GenAI device that explains why a specific patient may be at risk of a disease and prompts the HCP
to undertake further clinical review of the patient’s EHR record vs. a GenAI device that provides a direct diagnosis
for the HCP to rely on); (ii) the degree of personalization based on individual patient data; (iii) the context of
delivery (e.g., unsolicited alerts vs. user-initiated queries); (iv) the presence and effectiveness of human-in-theloop safeguards downstream of the output; and (v) the degree to which the user can independently assess the
basis for the output and verify its appropriateness. Risk may be mitigated through clear presentation of
alternatives and uncertainty, traceability to authoritative sources, appropriate human review, and technical
controls that support adherence to the intended use and predefined operating boundaries.

3. CDRH seeks input on whether and when a GenAI-enabled function that results in delivery of clinical
information to patients, as opposed to HCPs, could present different or higher risks, while also recognizing the
potential benefits associated with improved patient empowerment, engagement, and access to clinical
information. What device characteristics, output features, or safeguards might mitigate risks that could arise
when a user lacks the domain knowledge to independently evaluate an output, without unnecessarily
underestimating patient capability?

CPC encourages the Agency to include patient-facing GenAI-enabled device functions within any future revised
regulatory frameworks. In addition to the considerations covered in the Discussion Paper regarding the
importance of patient empowerment, engagement, and access offered by these technologies, the Agency should
additionally consider that by nature of existing regulatory and professional controls, patients themselves are
often practically limited in steps that they can materially take in exercising healthcare. An erroneous or
inappropriate patient-facing output from a GenAI-enabled device function may therefore be limited by the
subsequent lack of access to enacting the steps of treatment. For example, if a patient-facing GenAI-enabled
device may suggest that a particular surgical or pharmaceutical course of treatment should be undertaken, but

COMBINATION PRODUCTS COALITION PAGE 5 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

the patient must first undergo evaluation by a healthcare professional who determines whether such treatment
is clinically appropriate and can be prescribed or performed. The inherent reliance of patients on healthcare
practitioners should be considered in the GenAI-enabled device risk framework as to not over-regulate patient
access to these important technologies.

4. CDRH seeks input on whether and how the distinction between generalist and specialist physicians might
inform the assessment of risk for HCP-facing GenAI-enabled functions. Under what circumstances, if any, might
risk be affected when an HCP who lacks the relevant clinical specialist knowledge receives information that falls
within a specialist area of practice? What device characteristics or safeguards might mitigate such risks?

CPC recognizes that GenAI-enabled devices may help extend access to specialty knowledge across a broader
range of healthcare settings. However, the Agency should avoid imposing restrictions based solely on specialist
versus generalist status, as healthcare professionals are expected to exercise independent clinical judgment
consistent with existing practice-of-medicine principles.

It may be appropriate for the Agency to place recommendations in GenAI-enabled device labeling which cautions
against use by healthcare practitioners without appropriate knowledge or training, however, more significant
restrictions such as within formal intended use or indications for use statements are not advised.

5. For multi-turn conversational GenAI-enabled devices that may migrate from providing “non-directive”
information to “action-directing” information over the course of an exchange, how should risk be assessed across
realistic conversational trajectories? How could the intended use of such a device be characterized when its
behavior is emergent across a conversation?

CPC supports the Agency's consideration of conversational trajectories when assessing risk for multi-turn
conversational GenAI-enabled devices. CPC recommends that risk assessment for multi-turn conversational
GenAI-enabled devices be anchored to the intended use as declared by the sponsor, with technical controls
(e.g., topic guardrails, scope-detection classifiers, automatic session termination or handoff) used to prevent
conversational drift into medical device functions not covered by the authorization. The Agency may consider
evaluation of conversational trajectories, including cumulative context, topic drift, repeated prompting,
contradictory information, and attempts to move the system beyond its intended scope. The intended use should
clearly describe the functions the device is intended to perform and the limits of those functions as conversations
evolve. Additional clarity regarding how conversational context, topic drift, and evolving user interactions should
be considered during risk assessment would help sponsors develop appropriate evaluation approaches for multiturn conversational devices.

6. For GenAI-enabled care escalation functions, CDRH is considering how the evaluation may account for both
under-escalation and over-escalation. How could manufacturers characterize and weigh these two directions of
error, given that they may not be commensurable and that acceptable trade-offs may vary by clinical context?

CPC recommends that under-escalation and over-escalation errors be characterized separately rather than
collapsed into a composite metric, since their harms differ in kind (delayed treatment vs. resource burden, patient
anxiety, and erosion of trust) and are not directly commensurable. Sponsors should be able to justify clinically
appropriate acceptance criteria for each direction of error, calibrated to the specific clinical context, including the

COMBINATION PRODUCTS COALITION PAGE 6 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

relative importance of sensitivity and specificity for urgent versus lower-risk clinical conditions. For highconsequence, low-prevalence conditions (e.g., stroke, sepsis), asymmetric weighting favoring sensitivity may be
appropriate; for lower-acuity conditions, specificity may warrant greater weight. CPC encourages the Agency to
engage clinical/medical societies to help inform domain-specific reference expectations rather than relying on a
single uniform standard across clinical contexts.

Comments Regarding Discussion Paper Section V: A Competency-Based Approach for
Premarket Evaluation of GenAI-Enabled Devices

7. Is the competency-based approach described above, i.e., device benchmarking followed by clinical
confirmation, a useful and appropriate framework for evaluating GenAI-enabled devices?

CPC is supportive of the Agency’s thinking regarding the need for a fundamentally different construct for
evaluation of GenAI-enabled device function performance given the unique and non-deterministic nature of
underlying models. However, CPC believes that clinical confirmation may not always be a required step
depending on intended use and risk level of the GenAI-enabled device under evaluation. Additionally, the Agency
should consider if there are clinical simulation-based assessments that may be available rather than actual clinical
investigations that may be leveraged to reduce burden on investigators and institutions.

While CPC believes the proposed framework is generally broad enough to encompass a range of GenAI-enabled
medical device use cases, including precision medicine applications, the Discussion Paper provides relatively
few examples illustrating how the framework may apply to GenAI-enabled functions used for biomarker-guided
patient stratification and treatment eligibility determination, treatment selection and therapeutic decision support,
companion diagnostic interpretation and decision support, clinical trial enrollment eligibility assessment, and
patient identification and screening functions used to support clinical trial enrollment decisions. Additional
examples or discussion of these use cases would help stakeholders better understand how competency
assessment, clinical confirmation, and evidence expectations may be applied in these contexts, particularly where
outputs may directly influence therapeutic decision-making, treatment eligibility, or clinical trial participation.

A related and more fundamental clarification would assist sponsors developing diagnostic tools in these use
cases. The Discussion Paper defines foundation models as AI models trained on large datasets that are applicable
across a wide range of contexts and can serve as the basis for further models built through fine-tuning. It further
defines GenAI as a class of AI models that generate derived synthetic content. Many diagnostic functions relevant
to the use cases identified above, including biomarker prediction from digital pathology images, imaging-based
patient stratification, and trial enrollment eligibility scoring, are built on foundation model architectures but
produce discriminative or predictive outputs rather than synthetic content. These tools use large pre-trained
vision transformers or multimodal encoders as feature extraction backbones, with task-specific fine-tuning layers
producing a binary label, a continuous risk score, or a ranked patient list. By the paper's stated definition, such
devices may not qualify as GenAI-enabled devices, even though they are architecturally dependent on a
foundation model and share many of the same regulatory challenges the paper is designed to address: opacity
of the pre-trained representation, limited transparency into third-party pre-training data and methodology,
emergent capabilities beyond the fine-tuning scope, and performance degradation across deployment
environments.

COMBINATION PRODUCTS COALITION PAGE 7 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

CPC respectfully requests that CDRH clarify whether the competency-based framework and, specifically, the
proposed Foundation Model Device Master File mechanism are intended to apply to foundation model-derived
devices whose outputs are discriminative or predictive rather than generative. If such devices are considered
outside the scope of this Discussion Paper, CPC requests that CDRH indicate the regulatory pathway under
which their specific challenges, particularly third-party foundation model transparency, cross-site and crosspopulation generalizability, and calibration of probabilistic outputs used for clinical decision-making, will be
addressed, and whether the January 2025 draft guidance “AI-Enabled Device Software Functions: Lifecycle
Management and Marketing Submission Recommendations”4 is the intended home for such considerations.

This clarification is particularly consequential for diagnostic tools used in clinical trial contexts. A foundation
model-based patient identification or eligibility scoring function may produce a binary output, e.g. eligible or not
eligible, that is not generative in nature, yet directly determines clinical trial participation and, downstream, patient
access to investigational therapies. The regulatory stakes of such outputs are comparable to, or greater than,
many GenAI-enabled informational functions addressed in this paper. Sponsors would benefit from explicit
confirmation of which evaluation framework governs these devices, and whether the discussion around question
25 about Foundation Model MAF program provides a mechanism to document the pre-trained backbone on
which they depend.

Additionally, CPC supports the competency-based approach as a potential evidence generation framework for
GenAI-enabled devices. However, the Agency should clarify how competency assessment activities integrate
with established device quality systems. Specifically, the Agency should describe how benchmarking and clinical
confirmation relate to user needs, design inputs, risk management, risk controls, verification, validation,
traceability, postmarket monitoring, and change management processes. Such clarification would help ensure
that benchmarking and competency-assessment activities complement quality system requirements rather than
create duplicative evaluation processes.

8. How might the two-axis risk framework described in Section IV be considered within the competency-based
approach described above to help determine the level of evidence needed for a premarket submission?

CPC has no specific comments on this question at this time.

9. Would a benchmarking structure such as the one described above be likely to provide adequate evidence of
clinical knowledge, analytic capabilities, safety behavior, communication, and generalizability to support a
reasonable assurance of safety and effectiveness? Are there elements that are missing, redundant, or
inappropriately categorized? Are there externally developed standards that could be leveraged?

CPC finds the proposed benchmarking taxonomy (Safety, Clinical Proficiency, Generalizability, and Agentic
Capabilities) to be a reasonable starting framework. As the Agency further develops benchmarking approaches
for GenAI-enabled devices, CPC encourages the Agency to leverage externally developed standards and
frameworks where they exist and are fit for purpose. Examples may include ISO/IEC 42001 (AI Management
Systems), ISO/IEC 23894 (AI Risk Management), ISO 14971 (Medical Device Risk Management), and emerging
NIST AI Risk Management Framework resources, including NIST AI 600-1 (Generative AI Profile), where
Available at https://www.fda.gov/media/184856/download.

COMBINATION PRODUCTS COALITION PAGE 8 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

appropriate. Leveraging established standards, where appropriate, may promote consistency, reduce
unnecessary duplication, and provide greater predictability for sponsors evaluating GenAI-enabled devices. CPC
would also welcome additional insight into how the Agency views broader system considerations, including
interactions between models, software components, users, workflows, and external dependencies, within the
proposed benchmarking and evaluation framework.

CPC is broadly supportive of a benchmarking structure as the Agency has described, however, would like to
advise of a critical difference between human healthcare practitioners and FDA-regulated systems as it relates to
other regulated medical products which may provide inputs to the GenAI device function or may be surfaced as
a component of GenAI output. That consideration is the existing labeling and intended use of the “other”
regulated medical products. Specifically, as a matter of medical practice, a human healthcare practitioner may
determine that off-label use of a medical product is appropriate in the context of their treatment of an individual
patient. However, CPC does not believe that it would be appropriate for GenAI device functions to suggest the
use of a regulated medical product that is inconsistent with that product’s intended use and labeling. Therefore,
the Agency should consider how the proposed benchmarking structure can include challenges to ensure that
the GenAI device function does not provide outputs that are inconsistent with the approved use of other regulated
medical products.

Additionally, CPC considers the proposed benchmarking framework to be a reasonable foundation for assessing
GenAI-enabled devices. However, the Agency should recognize that the safety and effectiveness of a GenAIenabled device may depend not only on the performance of the underlying model, but also on the context in
which the device is deployed, including user interactions, clinical workflows, information sources, and other
safeguards implemented by the manufacturer. Benchmarking and clinical evaluation should therefore assess
device performance within representative conditions of use and intended deployment environments, as
appropriate and proportionate to risk. This would help ensure that evaluations reflect real-world performance
while maintaining a flexible and risk-based approach to oversight.

10. CDRH recognizes that publicly available benchmarking assets may be subject to data contamination,
saturation, and limited real-world representativeness. How should a sponsor establish that performance on a
given benchmark predicts safe and effective real-world behavior for the device’s intended use? What evidence
should support the construct validity of a benchmark used to gate device evaluation, and what role should
sponsor-developed benchmarks play given potential concerns around independence and optimization to the
test?

CPC has no specific comments on this question at this time.

11. CDRH is considering that clinical confirmation for a GenAI-enabled device might not require a prospective
clinical study in every case, and has described above a range of approaches of increasing rigor and patient
exposure. How might a sponsor select and justify a confirmation approach tailored to a device’s intended use
and proportionate to the device’s risk profile? Are there device types or risk profiles for which one or more of
these approaches would be insufficient or inappropriate? How might the anticipated distribution of real-world
inputs be taken into consideration? Are there other methods of clinical confirmation that might help inform the
evaluation of GenAI-enabled devices?

COMBINATION PRODUCTS COALITION PAGE 9 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

CPC supports the two-axis (activity vs. consequences) risk-proportionate approach to clinical confirmation for
GenAI-enabled devices. CPC recommends that the Agency consider factors such as: (i) the position of the device
on the two-axis risk framework; (ii) the availability and representativeness of retrospective real-world data or
synthetic data for the intended use; (iii) the reversibility of any action taken; and (iv) the extent to which
benchmarking has already characterized behavior in the deployment context. Given the diversity of GenAIenabled device functions, the Agency should maintain flexibility in the types of evidence that may be appropriate
to support clinical confirmation, allowing sponsors to justify approaches that are proportionate to the device's
intended use and risk profile.

12. How should sponsors achieve statistically meaningful performance measurement for GenAI-enabled devices?
Where synthetically generated inputs supplement real patient data, how should sponsors account for
differences between the synthetic and real-world distributions when estimating performance, and under what
conditions, if any, is it appropriate to combine benchmarking evidence and clinical confirmation evidence to
support a single performance estimate?

The Agency should provide additional clarity regarding benchmarking methodologies, including considerations
related to repeatability, reproducibility, benchmark validity, uncertainty, and the evaluation of non-deterministic
outputs, to support consistent and predictable implementation of a competency-based framework.

13. For which clinical domains, device functions, or subpopulations is synthetic data particularly well-suited, or
particularly inadequate, as a supplement to real-world evidence? What safeguards would mitigate the risk that
synthetic data generated by models of the same class as the device under evaluation reproduces the very
performance gaps the evaluation is intended to detect, particularly for underrepresented subgroups?

CPC supports the Agency’s openness to considering the use of synthetic data as a component of a competencybased evaluation of GenAI device functions and we believe that use of these data is particularly advantageous
where human patient data are difficult to obtain in statistically significant quantities or where it is ethically
problematic to collect such data. Examples include rare diseases and populations where it is difficult to obtain
informed consent.

14. For open-ended device outputs, how should performance comparators and acceptance criteria be selected?
When a panel of qualified clinicians serves as the comparator, how should the applicable standard (for example,
the standard of care versus the performance of a median clinician in practice) be defined and
justified? Should generalist or specialist physicians be used as a performance standard? When should humanAI team performance, rather than the device operating alone, serve as the basis for evaluation?

CPC supports the use of qualified clinician adjudicators as an appropriate comparator for many open-ended
GenAI outputs. Comparator selection should be aligned with the intended users, intended use, and clinical
context of the device. For more autonomous functions, standalone device performance may also be relevant.
Where clinician panels are used as comparators, reporting inter-rater reliability may help contextualize observed
device performance.

CPC notes that intended users may not always be limited to either generalists or specialists, and that many
GenAI-enabled devices may be used across a range of healthcare settings and levels of clinical expertise. The

COMBINATION PRODUCTS COALITION PAGE 10 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

Agency should maintain flexibility in comparator selection and allow sponsors to justify comparators that are
representative of the intended users and clinical context in which the device will be deployed.

Additionally, CPC supports the Agency providing additional clarity regarding the role of human factors and
usability engineering in the evaluation of GenAI-enabled devices. Clinical performance depends not only on
model output quality but also on how users understand, interpret, and act upon those outputs. Considerations
such as user comprehension, automation bias, critical task performance, and human-AI team effectiveness may
be important factors in demonstrating the safety and effectiveness of GenAI-enabled devices, particularly where
users are expected to exercise clinical judgement based on device outputs. Additional clarity from the Agency
in this area would support consistent and predictable evaluation of GenAI-enabled devices while ensuring that
assessments appropriately reflect the interaction between users and AI-generated outputs in real-world clinical
settings.

15. Are there ways in which performance might be assessed relative to the care, technology, or course of action
likely to occur in the absence of the device, rather than to the comparators described in this section? What
approaches might be used to identify and justify a comparator such as unaided clinical judgment, delayed
specialist review, or no intervention?

CPC has no specific comments on this question at this time.

16. Should independent third parties be involved in some or all aspects of a competency-based approach,
including device benchmarking and clinical confirmation? If so, in what ways might qualified, independent thirdparty participation contribute to this approach, and what qualifications and independence criteria should apply?
Are there aspects of the assessment for which third-party involvement would be impractical or inadvisable? What
safeguards or program design features would be critical to prevent third-party participation from limiting
competition or preventing innovation?

CPC has no specific comments on this question at this time.

17. CDRH is exploring whether a competency-based approach could be applied to devices that incorporate a
variety of underlying model architectures, such as multimodal vision-language models and generative or
predictive world models. Are there aspects of the described approach that might be ineffective or inapplicable to
these devices? What additional or distinct considerations might need to be addressed?

CPC supports the competency-based approach as a useful starting framework for evaluating GenAI-enabled
devices. However, the Discussion Paper provides relatively few examples illustrating how the proposed
benchmarking and competency assessment approaches may be applied to imaging-based GenAI-enabled device
functions. Additional discussion or examples would help stakeholders understand how the framework may be
adapted for devices whose performance depends on clinically relevant image features, image interpretation, or
other non-conversational inputs, rather than primarily text-based interactions.

COMBINATION PRODUCTS COALITION PAGE 11 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

Comments Regarding Discussion Paper Section VI: Postmarket Monitoring for GenAI-Enabled
Devices

18. 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. Under what
conditions might such an approach be appropriate, and what characteristics of a monitoring program would need
to be in place to justify reduced premarket evidence? Are there device types or risk profiles for which this
approach would not be appropriate?

CPC supports, in principle, the concept of accepting greater premarket uncertainty in exchange for robust
postmarket monitoring, provided such an approach remains risk-proportionate and is supported by appropriate
safeguards. Given the diversity of GenAI-enabled device functions and intended uses, CPC encourages the
Agency to provide additional clarity regarding the circumstances under which increased reliance on postmarket
monitoring may be appropriate, the types of monitoring activities that may be considered sufficient, and how
these considerations may vary based on device risk and intended use. Greater clarity in this area would support
consistent implementation and help sponsors develop appropriate evidence-generation strategies across the
total product lifecycle. In general, this approach may be more suitable to moderate to low risk devices on the
two-axis framework as opposed to devices with high-risk functions where an incorrect outcome could lead to
serious irreversible harm. Where the Agency considers greater reliance on postmarket monitoring, the Agency
should clarify how GenAI-specific monitoring activities relate to existing quality system and postmarket
surveillance processes, including complaint handling, adverse event management, CAPA, trend analysis, risk
management review, and field actions. Clear alignment with established processes would support consistent
implementation, maintain accountability, and help ensure postmarket monitoring activities complement existing
quality system requirements rather than create duplicative processes.

The Agency may also wish to provide additional clarity regarding factors that could support greater reliance on
postmarket monitoring, including the availability of infrastructure to support ongoing monitoring activities, the
sponsor's ability to identify and respond to emerging issues, and other characteristics that may contribute to
confidence in a postmarket monitoring program. Examples may include predefined approaches for monitoring
device performance, processes for identifying when further evaluation or corrective action may be appropriate,
and clear assignment of responsibilities for postmarket oversight.

19. Please comment on the potential approaches to postmarket performance evaluation, including periodic rebenchmarking, sample-based clinician review, and performance degradation monitoring. What additional
approaches should CDRH consider, and how should the cadence and triggering events for reassessment be
determined?

CPC has no specific comments on this question at this time.

20. Please comment on whether the proposed approaches to postmarket monitoring can be facilitated by
machine-based supervisory agents. What considerations, including the evaluation and reliability of the
supervisory agent itself, should CDRH take into account for such an approach?

COMBINATION PRODUCTS COALITION PAGE 12 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

CPC recognizes that machine-based supervisory agents may improve the scalability and timeliness of postmarket
monitoring activities for GenAI-enabled devices. However, for clinically consequential issues, such tools should
augment rather than completely replace appropriate human oversight. CPC encourages the Agency to provide
additional clarity regarding the role, evaluation, and reliability of supervisory agents used to support postmarket
monitoring, including how sponsors may demonstrate that these tools are appropriate for their intended
monitoring functions. Greater clarity would help support consistent implementation while maintaining confidence
in postmarket monitoring processes. Additional clarity regarding potential exceptions, including situations in
which greater reliance on machine-based supervisory agents may be appropriate, would help sponsors develop
monitoring approaches that are proportionate to device risk and intended use. As the Agency further considers
the use of machine-based supervisory agents, additional clarity regarding factors that may influence confidence
in their use would be beneficial. Examples may include the reliability of the supervisory agent for its intended
monitoring function, the data available to support monitoring activities, the potential impact of model changes
over time, the degree of independence between the supervisory agent and the system being monitored, and the
extent to which independent human review remains appropriate for clinically consequential findings.

21. What roles might clinicians, healthcare institutions, professional societies, standards-setting bodies, and
other stakeholders appropriately play in postmarket monitoring, and how can these roles be structured without
diffusing manufacturer accountability?

CPC has no specific comments on this question at this time.

22. CDRH envisions that a manufacturer’s premarket competency-based assessment might serve as a baseline
against which post-deployment modifications could be re-evaluated. How might the extent of re-benchmarking
or other evidence be scaled to the nature and expected impact of a given modification? For example, are there
categories of change that might not significantly affect safety or effectiveness of a GenAI-enabled device, or
might be appropriately managed within a sponsor’s quality management system versus requiring the FDA
premarket review and authorization before implementation? Are there categories of changes that might be
appropriate for inclusion in a PCCP?

CPC has no specific comments on this question at this time.

23. How might PCCP concepts or other change-control approaches be adapted for GenAI-enabled devices when
the nature or scope of future modifications cannot be fully prespecified?

CPC recognizes that modifications to foundation models, prompts, guardrails, retrieval systems, external tools,
tuning datasets, and other GenAI-enabled device components may affect device performance. The Agency should
provide additional clarity regarding how changes to these elements may be evaluated within existing change
management and PCCP frameworks, including considerations for determining when a modification may warrant
reassessment, re-benchmarking, or additional review. Such clarity would support consistent implementation and
help manufacturers effectively manage GenAI-enabled device modifications throughout the product lifecycle.

CPC encourages the Agency to recognize that effective AI Governance may be implemented through a variety of
organizational models and processes. As PCCP concepts evolve for GenAI-enabled devices, sponsors should
be permitted to reference established internal AI Governance frameworks and lifecycle management processes

COMBINATION PRODUCTS COALITION PAGE 13 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

as part of a PCCP submission. The Agency’s evaluation should focus on the adequacy of the sponsor’s AI
Governance approach for managing risk and implementing changes versus an Agency defined AI Governance
structure.

Additionally, CPC requests clarification on how GenAI lifecycle and change-management expectations will
interface with the Quality Management System Regulation and the combination-product CGMP requirements
under 21 CFR Part 4. The Agency should provide additional clarity regarding the elements of a GenAI-enabled
combination product that may be important to consider within change-management and quality-system
processes, including factors such as model or service versions, system prompts, retrieval sources, guardrails,
connected tools, software dependencies, and deployment environments. The Agency should also clarify how
changes to these elements may be considered across the device, drug or biologic, labeling, human factors, risk
management, and postmarket reporting processes. The Agency should provide additional insights into how
PCCPs may be used for device constituent parts of drug- or biologic-led combination products.

24. For devices built on third-party foundation models, changes to the underlying model may be initiated by the
third-party model developer rather than the device manufacturer. How can a manufacturer detect, evaluate, and
respond to such changes in a timely manner? What mechanisms—for example, contractual, technical, or through
a PCCP—could provide reasonable assurance that third-party developer-initiated changes do not compromise
the safety or effectiveness of the device?

CPC recognizes that manufacturers may have limited visibility into changes made to third-party foundation
models while remaining responsible for the safety and effectiveness of their devices. The Agency should provide
additional clarity regarding reasonable mechanisms manufacturers may use to identify, assess, and respond to
foundation model changes, including change-notification processes and other risk-based approaches that
support ongoing oversight of third-party dependencies.

CPC recognizes that manufacturers remain responsible for the safety and effectiveness of devices built upon
third-party foundation models, yet may have limited visibility into changes initiated by foundation model
developers. As discussed in question 25 below, voluntary Foundation Model MAFs may provide a useful
mechanism to support change management and communication throughout the product lifecycle.

CPC encourages the Agency to explore whether Foundation Model MAFs could facilitate change-notification
processes for model updates. Under such an approach, foundation model developers could provide change
notifications directly to the Agency through the MAF, and the Agency could then notify sponsors that have
referenced the applicable MAF in a marketing submission and provide sponsor-facing summaries containing
information necessary to assess the potential impact of the change while preserving proprietary information and
intellectual property. Where appropriate, sponsor-facing summaries of relevant changes could help
manufacturers assess the potential impact of model updates on their devices without requiring disclosure of
proprietary information from the foundation model developer.

Such an approach could improve transparency, support timely risk assessment by sponsors, and facilitate
effective management of third-party model dependencies while maintaining appropriate protection of intellectual
property and preserving manufacturer accountability for device performance.

COMBINATION PRODUCTS COALITION PAGE 14 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

CPC believes PCCPs may provide a useful mechanism for establishing, in advance, risk-based approaches for
managing anticipated Foundation Model changes. Such approaches could help facilitate consistency and
predictability regarding verification, validation, and any appropriate device requalification activities associated
with planned model updates, while maintaining flexibility across different Foundation Model technologies and
use cases.

Comments Regarding Discussion Paper Section VII: Other Topics

25. Would voluntary Foundation Model MAFs be practical for foundation model developers to provide and to
keep current, and what content might they include? Given that participation would be voluntary and that model
developers may have limited incentive to disclose safety-relevant information, what would make such a program
sufficiently useful for premarket review? Are there alternative mechanisms CDRH should consider for obtaining
information about underlying foundation models?

CPC is supportive of CDRH’s openness to voluntary Foundation Model MAF submissions under the existing
Device Master File program. Foundation Models have significant potential to transform biological insights into
clinically meaningful tools to accelerate scientific innovation, improve the interpretation of complex information,
and enhance patient care. A voluntary, confidential, and flexible MAF framework could improve the consistency
and efficiency of the Agency’s review by allowing Foundation Model developers to provide information that the
Agency review teams could reference across submissions. The framework should remain risk-based and
recognize that Foundation Models differ in architecture, training methods, intended uses, update practices, and
levels of autonomy. A Foundation Model MAF could describe the model’s architecture, capabilities and
limitations, training methods and data characteristics, verification and validation, cybersecurity, plans for updates
and monitoring, and other safety measures. Because participation is voluntary, sponsors should remain
responsible for independently evaluating the Foundation Model use in the context of their device validation to
demonstrate the safety and effectiveness of the end-to-end device that is built on or incorporates the Foundation
Model. CPC encourages the Agency to provide additional clarity regarding whether the voluntary Foundation
Model MAFs may be applicable to a broader range of AI architectures, including multimodal and discriminative
models, in addition to generative foundation models. Such clarification would help ensure that the MAF
framework remains adaptable as AI-enabled medical devices continue to evolve and would support more
consistent application across different model types and use cases.

In addition, CPC encourages the Agency to explore mechanisms that would allow sponsors who have been
authorized to reference a Foundation Model MAF to access sponsor-facing summaries of relevant information.
Such summaries could provide sufficient transparency regarding model characteristics, limitations, and update
history to support manufacturers in managing devices throughout the product lifecycle while preserving
confidential and proprietary information held within the MAF.

26. Are there additional considerations that inform the premarket and postmarket evaluation of agentic GenAIenabled devices, beyond those applicable to non agentic GenAI-enabled devices? How should the elevated risk
associated with autonomous multi-step action, tool use, and reduced opportunity for human review be reflected
in acceptance criteria and oversight?

COMBINATION PRODUCTS COALITION PAGE 15 OF 16 combinationproducts.com
Docket No. FDA-2026-N-7874

CPC supports the Agency's recognition that agentic AI-enabled devices may present different considerations
than non-agentic GenAI-enabled devices. Additional clarity regarding the characteristics that may influence the
risk profile of agentic GenAI-enabled devices would be valuable, including factors such as the degree of
autonomy, use of external tools, ability to initiate actions, level of human oversight, and potential impact of
downstream actions. Greater transparency regarding how the Agency may consider these characteristics within
its risk-based approach would help sponsors develop appropriate evidence-generation, monitoring, and riskmanagement strategies.

CPC recognizes that agentic AI-enabled devices may introduce additional cybersecurity considerations relating
to external tool use, prompt injection, model manipulation, and other autonomous behaviors. The Agency should
provide additional clarity regarding how existing cybersecurity principles and risk-management approaches may
apply to these emerging technologies and how such considerations may be addressed during the evaluation of
agentic AI-enabled devices. CPC further encourages the Agency to align any future expectations for agentic AIenabled devices with existing FDA medical device cybersecurity guidance, where appropriate, to promote
consistency, leverage established cybersecurity principles, and avoid the creation of duplicative or divergent
requirements.

***

CPC supports the further discussion on regulation of GenAI-enabled medical devices given their importance to
the future of healthcare and the need to balance FDA oversight with access for patients and providers. We
appreciate the opportunity to provide input on the Discussion Paper and welcome any questions or further
discussion.

Yours truly,

Bonnie Odom
On behalf of the Combination Products Coalition

COMBINATION PRODUCTS COALITION PAGE 16 OF 16 combinationproducts.com