WhaleTeq Co., Ltd.

WhaleTeq Co., Ltd. appreciates the opportunity to comment on FDA’s discussion paper on generative AI-enabled medical devices (Docket No. FDA-2026-N-7874).

Based on our experience in physiological signal testing and medical device verification, we would like to highlight four practical considerations for sensor-based and measurement-oriented AI-enabled medical devices:

1. Verification should cover not only the AI model, but also the relevant complete device pathway.
2. Real patient data and controlled simulation provide different but complementary verification evidence.
3. The value of a benchmark should be based on its credibility, traceability, and intended test purpose rather than dataset size alone.
4. Premarket benchmarks should, where practical, be designed so they can later serve as post-change regression baselines.

Our main view is that GenAI may increase the scale, variability, and frequency of verification, but the underlying regulatory principles remain the same: evidence should be appropriate to intended use and risk, each test method should be applied within its demonstrated limits, and relevant device changes should trigger risk-based re-verification.

Please see the attached document for our full comments, including the rationale, limitations of physiological simulation, and our responses to the relevant FDA discussion questions.

WhaleTeq Co., Ltd. | Docket No. FDA-2026-N-7874

Comments on Considerations for the Regulation of
Generative AI-Enabled Medical Devices

Submitted by WhaleTeq Co., Ltd.
Docket No. FDA-2026-N-7874
WhaleTeq Co., Ltd. appreciates the opportunity to comment on FDA's discussion paper on generative AI-enabled
medical devices. WhaleTeq develops technologies for physiological signal testing and medical device verification.
Our work primarily covers four layers: Simulator System; Clinical Data / Software-Defined Bio-signal (SDB);
Verification / Regression Software; and Customized Integration.
Our comments focus on sensor-based and measurement-oriented AI-enabled medical devices. We believe GenAI
may require new verification methods, while the underlying regulatory logic remains familiar: evidence should be
appropriate to intended use and risk, each test method should be applied within its demonstrated scope, and
changes that may affect safety or effectiveness should be re-evaluated.

1. Verification should cover the model and the complete device pathway
Relevant to Questions 7 and 9
Model-level benchmarking is important, but for a sensor-based medical device it may not be sufficient by itself.
The final clinical or measurement output can also depend on the sensor interface, signal acquisition, signal
processing, software, and interactions among these components.
Physiological input -> Sensor -> Signal processing -> Algorithm / AI -> Device output
The regulatory objective is to establish reasonable assurance that the medical device performs safely and
effectively for its intended use. Therefore, where relevant, model-level competency assessment should be
complemented by system-level verification of the device pathway.
Controlled physiological signals are useful because they provide known and repeatable inputs. Manufacturers can
verify how the device responds to a defined condition and repeat the same condition after a design change. Their
role is to verify the portion of device performance that can reasonably be evaluated under controlled input
conditions, while real-world clinical performance is addressed through appropriate clinical evidence.

2. Real patient data and simulation provide complementary verification evidence
Relevant to Questions 10, 11, 12 and 13
Real patient data and simulation answer different verification questions. Real patient data is important for
representing clinically observed physiological variability. Controlled simulation is useful for creating known and
repeatable conditions for robustness, boundary, stress and regression testing.
Real patient data supports clinical representativeness. Controlled simulation supports repeatability and
controlled coverage of defined test conditions.
The boundary should be explicit. A simulator represents a defined subset of physiological and physical conditions.
For example, in optical physiological sensing, performance may be affected by tissue characteristics, optical path,
sensor geometry, device-to-body coupling and other physical interactions. Playing back a clinically recorded
waveform represents the waveform, while those additional physical factors remain outside the scope of that
playback.
A simulated physiological signal therefore represents only the parameters and conditions that the simulation
method has been designed and characterized to reproduce. Its test purpose and limitations should be stated.
Simulation is most useful when a manufacturer needs to control a specific input, repeat the same test
consistently, explore defined edge or stress conditions, or reproduce the same test after a modification. Clinical
confirmation remains necessary when human anatomy, physiology, use conditions or other real-world factors
materially affect device performance.

September 8, 2026
 WhaleTeq Co., Ltd. | Docket No. FDA-2026-N-7874
3. Benchmark value should be based on credibility and intended test purpose
Relevant to Questions 10 and 12
The purpose of verification is to provide credible evidence for a defined requirement or risk. A medical-device
benchmark should therefore start with a clear verification question.
What is being tested? What does the input represent? Where did it come from? What is the reference or
ground truth? What result is expected? What are the limits of the test method? Can the test be
reproduced?
A smaller, well-characterized dataset may provide stronger evidence for a specific verification purpose than a
much larger dataset with unclear provenance, ground truth, intended use or limitations. We encourage FDA to
emphasize fitness for purpose, traceability, reproducibility and clearly stated limitations when considering
benchmark validity.

4. Premarket benchmarks should be designed as reusable post-change regression baselines
Relevant to Questions 19 and 22
AI-enabled devices may change more frequently than traditional medical devices, but the underlying changecontrol question is the same: what changed, what could be affected, and what evidence is needed to show that
the device remains safe and effective?
Where practical, premarket verification should be designed so that relevant tests can later be reused as a
regression baseline. A baseline may identify:
Device version + Test data version + Test condition + Expected result + Acceptance criteria + Actual
result
After a relevant software, firmware, algorithm, AI model or other device change, the affected tests can be
repeated under the same controlled conditions and compared with the established baseline. The scope of retesting should remain risk-based: a limited modification may require targeted regression testing, while a change
that may affect sensing, signal processing, algorithm behavior or clinical output may require broader reverification.
Repeatability is one of simulation's main strengths in this context. Its purpose is to determine whether a known
aspect of device behavior has changed when the same controlled challenge is applied before and after a
modification. Clinical and real-world evidence continue to address performance beyond the defined scope of the
simulated test.

Conclusion
We believe GenAI increases the scale, variability and frequency of verification, while the fundamental principles
remain the same: verification should cover the relevant device pathway; each test method should be used within
its demonstrated boundary; evidence should be fit for its intended verification purpose; and device changes
should trigger risk-based re-verification where appropriate.
For sensor-based AI-enabled devices, reusable verification infrastructure can support these principles through
four complementary capabilities:
Controlled physiological signal generation -> Clinical and software-defined test content -> Verification
and regression testing -> Device-specific integration
This infrastructure complements clinical confirmation. Its role is to make the controllable and repeatable portion of
device verification broader, more systematic and easier to reproduce throughout the product life cycle.
WhaleTeq appreciates FDA's consideration of these comments.
Respectfully submitted,
WhaleTeq Co., Ltd.
Taiwan

September 8, 2026