What You Should Know
- ATA Action, the advocacy affiliate of the American Telemedicine Association (ATA), formally submitted comments to Dr. Michelle Tarver, M.D., Ph.D., Director of the FDA’s Center for Devices and Radiological Health (CDRH), addressing Docket No. FDA-2026-N-7874 (Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion Paper and Request for Feedback).
- Endorses the FDA’s proposed two-axis risk framework—evaluating systems based on the clinical autonomy of the software and the potential severity of harm from an erroneous output—while urging the agency to regulate by clinical function and risk rather than by the underlying generative AI technology itself.
- Emphasizes the “Real-World Alternative” standard: argues that when FDA evaluates risk-benefit profiles for AI devices, it must compare performance against the care a patient would realistically receive without the tool—which in rural, low-income, and clinical shortage areas is often a multi-month wait or no medical care at all, rather than an idealized, immediate in-person physician visit.
Grounding Device Risk in Rural Reality
Central to ATA Action’s argument is an urgent call for the FDA to recalibrate how it conceptualizes clinical risk. In standard regulatory reviews, an investigational device is traditionally evaluated against an idealized clinical encounter led by an attentive physician.
In modern virtual care, however, that baseline is often a fiction. Across thousands of rural zip codes and underserved inner-city neighborhoods, the realistic alternative to an AI-supported screening or triage tool is not a prompt, high-touch consultation with a top academic specialist; it is a six-month waiting list, hours of travel, or no clinical care at all.
Joe Nye, Head of Federal Government Relations for ATA Action, stressed that regulatory evaluations must weigh this access vacuum: “AI has enormous potential to expand access to high-quality virtual care, but only if the regulatory framework is clear, proportionate, and built to last. FDA’s discussion paper is a strong start. We look forward to working with the agency to turn it into guidance that protects patients, gives developers and clinicians predictability, and gets safe, effective tools to the people who need them, especially in rural and underserved communities.”
To prevent algorithmic tools from perpetuating healthcare disparities, the group urged the agency to mandate that manufacturers validate models across real-world patient subgroups—explicitly testing for performance variations across rural populations, individuals with limited health literacy, and non-English-speaking patients, rather than relying exclusively on synthetic datasets.
A Two-Axis Framework: Regulating the Action, Not the Code
The trade group strongly endorsed the FDA’s proposed two-axis risk matrix, which evaluates systems based on two independent vectors: the degree of clinical autonomy the tool exercises and the potential severity of harm if an output is flawed.
Crucially, ATA Action insisted that oversight must regulate function and risk rather than the underlying technology. Labeling an entire software application as a high-risk medical device simply because it incorporates generative architecture or a conversational user interface threatens to sweep benign administrative or wellness tools into grueling pre-market clearance cycles. The group petitioned the agency to reaffirm its January 2026 clinical decision support (CDS) and general wellness guidances, clarifying that adding a natural-language chatbot interface does not automatically convert an excluded informational software function into an FDA-regulated medical device.
Where a platform performs multiple clinical tasks within a single patient encounter—such as drafting a visit summary while concurrently flagging an abnormal lab trend—the FDA should evaluate each function according to its specific risk profile rather than forcing the entire platform into a duplicative, fragmented regulatory clearance process.
The Preemption Crisis: Proposing a Federal-State Sandbox
While the FDA refines its medical device purview, state medical licensing boards and state legislatures are independently crafting disjointed rules regarding clinical AI, medical malpractice, and scope of practice. This is rapidly creating a chaotic regulatory patchwork where a federally authorized algorithm might be embraced in one state but effectively banned or restricted as the “unauthorized practice of medicine” just across state lines.
To head off this jurisdictional clash, ATA Action proposed creating a voluntary federal-state clinical AI regulatory sandbox. Modeled on the FDA’s voluntary TEMPO pilot (Technology-Enabled Meaningful Patient Outcomes) and built in direct coordination with the Department of Health and Human Services (HHS), the initiative would allow participating states to align with federal standards.
Under the proposed sandbox, developers would deploy semi-autonomous tools under phased clinician oversight and shared reporting standards, while the FDA exercises conditional, time-limited enforcement discretion. This coordinated environment would generate real-world clinical evidence without penalizing multi-state telehealth providers attempting to serve cross-border patient panels.
Pragmatic Transparency and Payer Alignment
The filing also cautions against regulatory over-engineering in clinical disclosures and data security:
- Proportionate Labeling: While patient-facing bots must disclose that a user is conversing with an algorithm and must never masquerade as a human clinician, warnings should be concise and plain-spoken to avoid “alert fatigue,” always providing a clear escalation path to human clinicians or emergency dispatch.
- Existing Privacy Rails: ATA Action urged the agency to rely on mature HIPAA and federal cybersecurity baselines rather than inventing generative-AI-specific mandates that introduce conflicting compliance regimes.
- Closing the Coverage Loop: Finally, the group pressed the FDA to formalize coordination with the Centers for Medicare & Medicaid Services (CMS). Without guaranteed coverage and sustainable reimbursement pathways from CMS and commercial payers, even the most rigorously validated clinical algorithms will fail to reach the vulnerable patient populations they were designed to serve.
