Hearing Industries Association (Bridget Dobyan)

Please see the attached comments submitted by the Hearing Industries Association.

October 19, 2026

U.S. Food and Drug Administration
Centers for Devices and Radiological Health
10903 New Hampshire Avenue
Silver Spring, MD 20993

Submitted electronically via Regulations.gov

RE: Docket No. FDA-2026-N-7874, Considerations for the Regulation of Generative AIEnabled Medical Devices: Discussion Paper and Request for Feedback

The Hearing Industries Association (HIA) appreciates the opportunity to submit these comments
in response to the Food and Drug Administration (FDA) regarding the Center for Devices and
Radiological Health (CDRH) discussion paper, Considerations for the Regulation of Generative
AI-Enabled Medical Devices: Discussion Paper and Request for Feedback, Docket No. FDA2026-N-7874 (Discussion Paper).

HIA is the national organization of manufacturers, suppliers, and distributors of hearing aids,
implants, assistive listening devices, components, parts, and power sources. Together, these
companies drive innovation and produced over 5 million hearing aids purchased in the United
States last year. Given the engagement of our members with hearing health professionals and
consumers, HIA’s mission is to be the trusted voice on product innovation, patient safety and
education, and public policy.

HIA supports FDA’s effort to engage stakeholders early regarding the regulatory considerations
associated with generative artificial intelligence (GenAI)-enabled medical devices. Artificial
intelligence (AI) advances offer significant opportunities to improve patient outcomes and
support clinical efficiency and proficiency, but they may also pose distinct risks, particularly in
diagnostic and therapeutic applications. Therefore, HIA recommends that FDA adopt a riskbased approach that prioritizes patient safety while providing competent, reasonable standards
and regulatory clarity. As part of this approach, we respectfully caution against applying
regulatory concepts developed for high-risk GenAI systems to conventional, bounded machinelearning functions. We also submit that low-risk GenAI functionality in an associated application
should not independently trigger a premarket submission when it does not change intended use,
enable autonomous clinical decision-making, or materially increase risk of a device.
I. Machine Learning in Hearing Aids

Like medical devices, AI is not homogeneous. Under the broad concept of AI, functionality and
limitations determine application. For example, hearing aids utilize machine learning and
incorporate deep neural networks (DNN) to improve speech-in-noise, speech intelligibility and
background noise reduction, and to make automatic environmental adjustments that improve user
experience. Whereas GenAI creates new content, DNN models classify existing data across
multiple layers and make predictions based on that existing data.

For example, hearing aids that utilize DNN models operate on top of baseline algorithms that
support safe performance. In many current hearing aid implementations, DNN-based functions
operate within bounded processing architectures and are supplemented by conventional signalprocessing controls, output-limiting mechanisms, and other risk controls. These functions are
based on complex pattern understanding and application; they are not based on the creation of
new content or directed action.

While this example falls outside the scope of the Discussion Paper, it provides useful insight into
the varying intended uses and risks of AI-related technologies and highlights the importance of a
regulatory framework that is sufficiently proportionate and responsive to technological
advancements and evolving models, rather than a one-size-fits-all approach. Such a structure
should recognize distinctions between intended uses and fundamental risk profiles, enabling
oversight that supports innovation while maintaining safeguards for patient safety. Therefore, we
respectfully submit comments encouraging FDA to take a risk-based approach to regulating
GenAI.

II. Applying a Risk-Based Approach to Classifying and Determining Requirements
for GenAI-Enabled Devices

As FDA contemplates GenAI-enabled medical devices, we encourage critical consideration of
the model’s safety and risk factors, intended use, and appropriate competency standards. Thus,
we support a risk-based approach to classifying and determining regulatory requirements for
GenAI-enabled medical devices.

Such an approach aligns with existing medical device regulatory frameworks; for example, even
though DNN models have safety functions built in, FDA already employs risk-based approaches
to regulating hearing aids that may be appropriate for other functions of AI, like GenAI-enabled
devices. Such oversight has kept hearing aids utilizing DNN models safe and effective for
patients for years.
Under the discussion questions relating to the two-axis framework FDA outlines, CDRH seeks
input on the delivery of information to patients (and resulting actions) as well as the role of
specialized knowledge of the healthcare professional in receiving that information. Hearing aid
technology highlights the low-risk execution of a machine-learning component, incorporating
failsafe mechanisms and professional oversight to support the overall safety and effectiveness of
these medical devices. Outside of the device, hearing care applications may utilize GenAIenabled functions to provide informational, educational, troubleshooting, communication, or
workflow-support capability without altering the intended use or fundamental risk profile of the
underlying device. Human oversight remains present, and outputs are advisory rather than
autonomous. Regulation of future or emerging technology should likewise consider the intended
use of the output, the individual action and consequences, and safeguards in place.

III. Types of Valid Scientific Evidence for the Evaluation of the Safety and
Effectiveness of GenAI-Enabled Devices Across the Total Product Life Cycle
(TPLC)

The Discussion Paper asks for feedback regarding the clinical domains, device functions, or
subpopulations for which synthetic data may be particularly well-suited as a supplement to realworld evidence, as well as the safeguards that would mitigate the risk that synthetic data
reproduces performance gaps (see question 13). We believe existing regulatory frameworks may
prove instructive. For example, under current FDA and Federal Trade Commission (FTC)
standards, medical device manufacturers are required to substantiate any claim made in product
materials. While all claims require substantiation, certain claims require a more rigorous basis
depending on the nature of the claim, the intended use of the device, and the potential
consequences if the claim is inaccurate or misunderstood. To illustrate, claims of clinical
performance generally require evidence generated through appropriately designed clinical
evaluation, which may include clinical investigation with human subjects where necessary to
demonstrate safety, effectiveness, and real-world performance. By contrast, lower-risk
functionality claims may be supported by non-clinical testing or other competent and reliable
evidence tailored to the claim being made.

This existing substantiation framework could provide an appropriate roadmap as FDA considers
GenAI implementation. Evidence expectations should be tied to the particular GenAI-enabled
function, intended use, the degree to which the output is informational, action-directing, or
action-taking, and the severity of harm that could result from reliance on an incorrect output.
Such an approach would help ensure that high-risk claims and functions are supported by robust
clinical evidence while avoiding unnecessary burdens for lower-risk features that do not
diagnose, treat, direct clinical action, or control safety-critical device performance. In all cases,
the evidence standard should remain tied to the claim, intended use, and potential consequences
of an incorrect output.
With respect to FDA’s proposal to shift to a post-market oversight model for GenAI-enabled
(and potentially broader regulation of AI) medical devices (see question 18), it is important to
consider existing limited review capacity and the impact that may have on a surveillance model.
Indeed, HIA raises concerns that shifting review considerations to post-market surveillance
might not be effective due to the Agency’s limited resources, and encourages a proportionate
framework for monitoring and traceability expectations based on function and risk.

IV. Conclusion

As discussed in these comments, varied, open-ended outputs may pose different risks. However,
there are applications (e.g., DNN in hearing aid devices) that provide built-in control
mechanisms and do not pose the same level of risk as GenAI-enabled devices. Where a device
would otherwise be exempt, the incorporation of low-risk GenAI-enabled functionality in an
associated application should not independently create a premarket submission requirement,
provided the function does not introduce a new intended use, autonomous clinical decisionmaking, or materially increased risk.

HIA appreciates the opportunity to comment and welcomes continued engagement with CDRH
as relevant agencies consider future regulatory approaches for GenAI-enabled medical devices.
We would be pleased to provide additional information, technical input, or perspective as FDA
continues this important work.

Sincerely,

Bridget Dobyan
Executive Director
Hearing Industries Association

1301 K Street NW, Suite 300W, Washington, D.C. 20005
(202) 975-0905 BetterHearing.org