Salvo Health (Jeff Glueck)

Please see attached comments from Salvo Health.

Salvo Health Comments – FDA GenAI Devices (FDA-2026-N-7874)

Salvo Health Comments – FDA GenAI
Devices (FDA-2026-N-7874)
Oct 19, 2026 · ​@Amanda Sussex

October 19, 2026

Dr. Michelle Tarver Director, Center for Devices and Radiological Health Food and Drug
Administration 10903 New Hampshire Avenue, Silver Spring, MD 20993

Dr. Rick Abramson Director, Digital Health Center of Excellence Center for Devices and
Radiological Health Food and Drug Administration 10903 New Hampshire Avenue, Silver
Spring, MD 20993

RE: Considerations for the Regulation of Generative AI-Enabled Medical Devices:
Discussion Paper (Docket No. FDA-2026-N-7874)

Dear Dr. Tarver and Dr. Abramson,

Salvo Health is pleased to provide feedback on the Considerations for the Regulation of
Generative AI-Enabled Medical Devices discussion paper. As a technology-enabled
provider, Salvo sees firsthand how regulatory clarity, or its absence, determines whether
technology can expand access to care. FDA's approach will shape whether providers like
Salvo can responsibly extend our licensed clinicians' reach to the patients who need it.

Provider shortages are reducing patient access and affordability across the country. At
the same time, patients increasingly expect care on demand, at their fingertips, and many
disengage from the health system when it cannot meet their needs promptly. With
clinicians burning out and too few licensed providers to meet Americans' needs,
technology and AI-assisted care must help bridge the gap.

Patient chat is the right place to start. It is high-touch, and it keeps patients from turning
to sources that are not clinically informed or evidence-based when they have a question
between visits. It is also highly trainable, which supports high-quality information,
consistent escalation, and easy handoff to a licensed clinician whenever a human is
needed.

Salvo Health is a technology-enabled provider of virtual, multidisciplinary care for people
with chronic gastroenterology, hepatology, and metabolic conditions. We have built a
technology platform to support patients between visits, yet nearly all of our care,

Page 1 of 6
Salvo Health Comments – FDA GenAI Devices (FDA-2026-N-7874)

including every patient chat message, is still delivered by licensed clinicians. Today,
licensed nurses, registered dietitians, and master's-level behavioral health clinicians,
answer patient chat messages about symptoms, nutrition, and coping.

We operate this way because there is no clear regulatory or reimbursable pathway for AIdelivered patient chat, not because the technology is the limiting factor. Without a
defined FDA framework, payers and CMS have little basis for recognizing AI-enabled care,
so models like ours default to licensed-clinician time for every interaction. This limits how
many patients we can serve and keeps care more expensive than it needs to be.

We are now evaluating generative AI for patient chat and intake triage, and we appreciate
the opportunity to comment as an operator of the clinical service these devices would
support. Salvo serves on the board of the Alliance for Connected Care and supports the
comments the Alliance submitted. We offer the following additional perspective from a
provider that would deploy these technologies directly with patients, focusing on the
questions most relevant to that experience.

Our overarching position is that FDA should create a clear, risk-proportionate pathway for
AI-delivered patient chat and intake triage within established chronic care, including a
path to graduated autonomy. A defined pathway would also give payers and CMS the
shared foundation they need to recognize this care.

In advance of our detailed comments, Salvo:

Supports a pathway to graduated autonomy. Providers should be able to move from
clinician review of AI-drafted responses, to sampled review, to independent operation,
function by function, as real-world evidence accumulates. This mirrors the supervisedto-independent progression the discussion paper cites for clinician credentialing, and
should be an available pathway rather than a mandatory starting point.
Urges FDA to weigh the established care relationship. Patient-facing chat delivered
within a clinician-led care plan, with a clear path to a licensed clinician, should not be
presumed high risk.
Recommends a guaranteed escalation path as a core safeguard. Patients should
always be able to reach a licensed provider with whom they have an existing care
relationship, either during normal business hours or by request for follow-up during
those hours. Functions that meet this standard should qualify for lower-risk treatment.
Encourages flexible, real-world evidence pathways. Retrospective review of cliniciandelivered chat, shadow deployment, and clinician corrections of AI drafts should count
toward both premarket and postmarket evidence.

Page 2 of 6
Salvo Health Comments – FDA GenAI Devices (FDA-2026-N-7874)

Urges FDA to compare performance to care without the device. For many patients,
the realistic alternative is a delayed reply, an unverified online source, or an ED visit.
Supports asymmetric escalation standards. Missed red flags should carry near-zero
tolerance, while over-escalation should be measured against the performance of
today's licensed clinicians.

Salvo offers the following detailed comments on how FDA's proposed framework would
apply to patient chat and intake triage.

Assessing Risk for Patient Chat and Intake Triage
Salvo supports the two-axis framework and recommends that FDA treat the availability of
downstream safeguards as an explicit risk modifier. A chat function backed by a staffed,
licensed escalation path with a defined response time presents meaningfully less risk
than the same function with no human behind it.

We recommend that FDA define a baseline escalation standard for patient chat. Patients
should always have access to a licensed provider with whom they have an existing care
relationship, either during normal business hours or by request for follow-up during those
hours. A function that meets this standard, within an established care relationship, should
qualify for lower-risk treatment under the framework.

Reversibility should also count. In our experience, most nutrition and symptommanagement guidance can be adjusted at the next interaction if it proves unhelpful. By
contrast, a missed red flag such as GI bleeding cannot be undone, and the framework
should reflect that difference.

We agree that directiveness should be judged on substance rather than wording. Much of
what Salvo's clinicians do in chat is help patients apply a plan their care team has already
set, such as adjusting a meal within a dietitian's plan or using a coping technique a
therapist has taught.

Guidance that operates inside a clinician-established care plan should be treated as less
directive than a new clinical recommendation. Worked examples drawn from chroniccondition management would give providers the predictability they need to design and
classify these functions.

Patient-facing outputs should not be presumed high risk. Salvo's patients are enrolled in
an established, clinician-led care plan, and they know how to reach their care team. That
relationship changes the risk profile of a chat interaction compared with an anonymous
consumer chatbot.

Page 3 of 6
Salvo Health Comments – FDA GenAI Devices (FDA-2026-N-7874)

Effective mitigations include clear disclosure that the patient is interacting with AI, onetap escalation to a licensed clinician, and outputs tested across health literacy levels. As
the Alliance notes, patients routinely make sound decisions about when they need a
human, provided they have the information and a clear pathway to do so.

Salvo supports assessing risk across realistic conversations rather than single messages.
Patient chat conversations often move from general questions to personal ones, and our
clinicians manage that drift every day.

A conversational device's intended use can be defined by its permitted scope plus
mandatory handoff triggers. Sponsors can then test it with multi-turn simulations
designed to provoke drift, including patients who minimize symptoms or push for advice
outside scope.

For care escalation functions, both directions of error matter, but they are not equal.
Under-escalation of red flags, such as GI bleeding, severe abdominal pain, or suicidal
ideation, should carry near-zero tolerance.

Over-escalation should be measured against the baseline of today's licensed clinicians,
for example their rates of urgent care and ED referral for similar messages. Unnecessary
ED use harms patients and the health system, so FDA should allow sponsors to set
different thresholds for different symptom categories rather than one standard for all.

Premarket Evidence for Patient Chat
Because licensed clinicians deliver nearly all of Salvo's chat today, we hold a large set of
real patient messages paired with licensed-clinician responses. Providers like Salvo are
well positioned to generate evidence through retrospective evaluation on these records
and through shadow deployment, where AI drafts responses alongside our clinicians
without reaching patients.

These approaches should suffice for lower-consequence functions such as nutrition
guidance within an established plan. Prospective studies remain appropriate for higherconsequence functions, such as symptom triage that could delay emergency care.
Evidence generated in normal clinical practice should count, consistent with the Alliance's
call for flexible, real-world evidence pathways.

Performance comparators should match the role being performed. For supervised
deployment, the comparator should be the performance of the clinician-AI team within
the intended workflow. For autonomous deployment, it should be the licensed clinician
who performs the function today.

Page 4 of 6
Salvo Health Comments – FDA GenAI Devices (FDA-2026-N-7874)

At Salvo, that means a registered nurse for symptom questions, a registered dietitian for
nutrition, and a licensed therapist for coping support. Comparing these functions to a
physician panel would set a standard that does not reflect how this care is actually
delivered.

Salvo strongly supports measuring performance against what patients experience in the
absence of the device. Even with a fully staffed care team, a patient may wait hours for a
reply, and many patients without access to specialty care have no one to ask at all.

In those moments, patients often search online, rely on unverified sources, or go to the
ED. Faster response, broader access, and reduced unnecessary utilization should count
as benefits in FDA's benefit-risk determination. An imperfect capability must be weighed
against a patient who cannot get care.

Post-market Monitoring of Patient Chat
Salvo supports greater reliance on post-market monitoring for lower-consequence
functions. This trade is appropriate when the sponsor runs continuous sampled clinician
review, tracks missed escalations, and sets rollback triggers in advance.

It is not appropriate for autonomous, high-consequence functions at launch. Those
functions should earn reduced oversight through the graduated pathway described
above.

We agree with the Alliance that clinician review and correction of AI outputs should be
recognized as a distinct monitoring approach. When Salvo's clinicians edit, reject, or
escalate beyond an AI-drafted response, each action is a direct signal of quality.
Systematically capturing these signals creates a continuous, clinically grounded record of
performance.

Like most providers, Salvo would build on third-party foundation models rather than
develop our own. Manufacturers should run a pinned model version and re-benchmark
before adopting a vendor update, supported by contractual notice obligations.

FDA should accept PCCPs that cover vendor model updates validated against the original
benchmark.

Salvo has built the technology to extend our licensed clinicians' reach, but today that
technology cannot be fully realized. Without clear standards for AI-delivered chat, there is
no path to the reimbursement and payer confidence that would let providers like us use it
at scale. Patients pay the price in longer waits, higher costs, and care that never reaches
them.

Page 5 of 6
Salvo Health Comments – FDA GenAI Devices (FDA-2026-N-7874)

Clear, risk-proportionate evidence standards and a defined pathway for graduated
autonomy would let providers expand access to licensed-quality guidance safely. We also
encourage FDA to coordinate with states, several of which now restrict AI in behavioral
health, so that providers operating nationally face consistent expectations.

We also encourage FDA to continue coordinating with CMS, including through its TEMPO
pilot alongside the CMS ACCESS model. Aligning FDA's framework with outcome-aligned
payment models would give providers a financial path to adopt AI-enabled chat that
reduces reliance on fee-for-service care.

Salvo appreciates FDA's consideration of these comments and welcomes continued
engagement with CDRH and the Digital Health Center of Excellence.

Respectfully submitted,

Jeff Glueck Chief Executive Officer, Salvo Health

Amanda Sussex SVP, Business Development and Partnerships, Salvo Health

Page 6 of 6