Wearlinq Inc. (Konrad Morzkowski)

Please see attached comments from Wearlinq.

wearlinq.com

October 18, 2026

Submitted electronically via Regulations.gov

Re: Docket No. FDA-2026-N-7874, Considerations for the Regulation of Generative AI-Enabled
Medical Devices: Discussion Paper and Request for Feedback
Dear Digital Health Center of Excellence team,
Wearlinq, Inc. appreciates the opportunity to comment on the Center for Devices and Radiological
Health's discussion paper on the regulation of generative AI (GenAI)-enabled medical devices. We
commend the Center for seeking stakeholder input early, before proposing policy, and we support the
two-axis risk framework and the competency-based approach to premarket evaluation.
About Wearlinq. Wearlinq makes the eWave, a wireless, continuous, FDA-cleared 6-lead ECG monitor,
and operates an independent diagnostic testing facility (IDTF) that provides continuous ECG monitoring
services. We have a commercial interest in how FDA treats GenAI functions that build on cleared devices.
Our comments focus on a question the discussion paper does not address directly: how GenAI can be
added as an interface to a device that is already FDA-cleared. Many cleared devices have two simple
interfaces, one that collects information from users and one that presents results to them. GenAI can
improve both without changing what the device measures, calculates or concludes. We address three
topics:
1. GenAI as an input interface that collects the same information a cleared device already collects.
2. GenAI as an output interface that lets users ask questions about results a cleared device already
produces.
3. Recognizing clinical review and source traceability as factors that lower risk, so cleared, supervised
devices can do at least as much as uncleared products.

1. GenAI as an input interface to a cleared device
Many cleared devices collect information from users through fixed interfaces such as buttons, forms or
diaries. In ambulatory monitoring, for example, patients log symptoms so physicians can match them to
the recorded data. A GenAI interface can collect the same information in natural language and convert it
into the same structured fields, often more completely, while the device's measurements, algorithms and
outputs stay the same.
The Agency has already cleared a device in which a conversational interface collects patient-reported
information and passes structured fields to deterministic, provider-configured logic (K253281). That
clearance came through a new 510(k). For a device already on the market, it is less clear whether
replacing a fixed input with a GenAI interface requires a new submission. The software change guidance
asks whether a change introduces or modifies a risk, and a GenAI interface introduces one specific,
testable risk: information extracted incorrectly. A strict reading could require a new submission even when
the structured inputs and every downstream function are unchanged.
We recommend the Agency clarify that a GenAI input interface may be added to a cleared device and
documented within the manufacturer's quality management system, without a new marketing submission,
when it meets verifiable conditions (Questions 2, 3, 5 and 22):
• Equivalence. It produces the same structured fields as the existing interface, with accuracy at least
equal, shown through benchmarking against the paper's elements for information gathering (E.2),
quantitative analysis (E.3), robustness (R.1) and subgroup performance (R.2).
• Traceability. Each structured entry links to the user's own words, and the user confirms it.

Wearlinq, Inc. comment, Docket No. FDA-2026-N-7874 | 1
 • Bounded outputs. The interface only asks, confirms and records, from a closed set of outputs. It
does not interpret device data or give advice, so conversational drift can be tested against that set.
• Deterministic escalation. Answers that signal urgent symptoms trigger fixed escalation through
rules outside the model (element S.1).
• Risk management and usability. The manufacturer documents a risk analysis for the new
interface, completes usability testing with intended users, and updates labeling and instructions for
use.
• Monitoring. Conversations are logged, and qualified reviewers check a sample on a defined
schedule, similar to the sample-based clinician review the paper describes for postmarket
monitoring (Question 19).
This keeps such changes in a defined quality-system lane, consistent with the documentation-only
approach in the Agency's guidance on deciding when to submit a 510(k) for a software change. We
recommend reserving PCCPs and new 510(k)s for changes outside this lane, such as a GenAI function
that alters the device's measurements, algorithms or results, or changes its intended use or users.

2. GenAI as an output interface to a cleared device
Cleared devices present results in fixed formats: reports, values and, for some patient-facing devices,
predefined result messages written into the device labeling. A GenAI interface can let users ask questions
about those results in natural language. For example, a physician could ask for a measurement the
device already calculates for a specific date and receive the cleared value, with a reference to where it
appears in the report, instead of searching through the report.

Figure 1. GenAI input and output interfaces around an unchanged cleared device.

Section 520(o)(1)(D) of the FD&C Act excludes from the device definition software that displays device
data and results, together with general information about those results, unless it is intended to interpret or
analyze them. It is not clear how this applies to GenAI that retrieves and restates cleared results in
natural language. Another commenter (comment 0126) distinguishes functions that present a cleared
output from functions that generate new clinical claims. We agree, and recommend the Agency apply that
line here (Questions 1, 2, 3 and 9):
• Clarify that a GenAI interface that only retrieves and restates cleared results, links each value to its
source, uses approved explanatory content and adds no new interpretation, is a display function
under section 520(o)(1)(D).
• Where such an interface is part of the device, allow the manufacturer to define its boundaries in the
marketing submission or a PCCP: the cleared results it can draw on, the approved explanatory
content, the permitted question types and the intended users. Validation would cover numeric
accuracy (E.3), declining questions that require new interpretation (S.2), uncertainty (S.3) and user
comprehension (E.4).
• Keep intended users as cleared for device functions. For patient-facing explanations of finalized
results that patients already have, such as a report their physician has released, clarify how a

Wearlinq, Inc. comment, Docket No. FDA-2026-N-7874 | 2
 device manufacturer may offer the same kind of explanation that EHR patient portals already
provide.
For GenAI functions that are part of the device, the underlying language models will change more often
than the rest of the device. We recommend the Agency allow PCCPs that use the benchmarks above as
the gate for model updates, and offer a streamlined way, such as a Special 510(k), to add such a PCCP
to a device already on the market (Questions 22, 23 and 24).

3. Clinical review and source traceability should lower risk (Question 1)
We support adding the availability of downstream safeguards and the traceability of outputs to the twoaxis framework. In an IDTF model, device outputs are reviewed by qualified technicians and then by the
ordering physician before anyone acts on them. A GenAI function whose outputs pass through this
review, and whose outputs can each be checked against a cited source, carries less risk than the same
function used without it. We encourage the Agency to recognize established clinical review, together with
source traceability, as factors that lower a function's position on both axes.
Regulatory treatment should depend on a function's intended use and the safeguards around it, whoever
deploys it. Wellness products and general-purpose chatbots already let people describe symptoms and
ask questions about their health data in natural language. EHR systems have added GenAI features that
answer clinicians' questions about a patient's record with cited sources and explain test results to patients
in plain language. Health systems report that these features save clinicians time and help patients get
answers sooner, and we support that progress. A cleared device with clinical review behind it should be
able to deliver the same benefits. If the same interface is harder to add to a cleared device than to an
uncleared product, the framework discourages companies from seeking clearance.
The discussion paper notes the argument that GenAI-enabled devices need accountability mechanisms
analogous to the professional and legal consequences clinicians face. Manufacturers of cleared devices
already operate under such mechanisms, including quality management system requirements, adverse
event reporting, recall reporting and FDA inspection (21 CFR Parts 820, 803 and 806). Applied
consistently, these factors would ensure that manufacturers of cleared, clinically supervised devices can
adopt GenAI interfaces on a path no more burdensome than uncleared products that offer similar
features.
Wearlinq appreciates the Agency's early engagement on this topic. Clear rules for adding GenAI
interfaces to cleared devices would support the Agency's goal of a regulatory approach that "allows
patients timely access to safe and effective medical devices." Patients could report symptoms in their own
words when they occur, including patients who find fixed forms hard to use. Clinicians would receive more
complete information and faster answers from results they already rely on. These improvements would
come from products built on validated measurements and clinical review, with the guardrails described
above. We would welcome the opportunity to discuss these comments or to participate in pilots of the
competency-based approach.
Respectfully submitted,

Konrad Morzkowski
CEO, Wearlinq, Inc.

Wearlinq, Inc. comment, Docket No. FDA-2026-N-7874 | 3