Federal health regulators are quietly working to keep their rules from colliding with a growing number of state programs that let AI perform clinical work, including writing prescriptions. STAT reported the effort on October 30, and the fight it describes is now live. The core question behind FDA AI prescribing rules is simple to ask and hard to answer: who is in charge when software, not a doctor, makes the clinical call?
STAT's reporting says officials at the Food and Drug Administration and the Centers for Medicare & Medicaid Services are working behind the scenes to head off a clash with states. The stated worry is that overlapping jurisdiction could stall adoption. STAT put it this way: “confusion over overlapping jurisdictions could stall health AI adoption.” No named official is quoted in the coverage I could access, and neither agency has publicly confirmed a specific coordination plan.
The timing matters. More states are standing up clinical-AI testing programs, so the jurisdictional question is no longer theoretical for companies, clinicians or patients.
FDA AI prescribing rules and the jurisdiction gap
The core problem is structural. States license clinicians and regulate the practice of medicine. The FDA regulates medical devices. An AI system that writes prescriptions without a clinician in the loop falls under both frameworks, which is why neither rulebook fits cleanly.
STAT's reporting goes further on one point. If AI performs clinical tasks without human supervision, those products would likely fall under FDA jurisdiction. That is a meaningful signal, because it would bring autonomous clinical software into the device regime. The sources I accessed do not say whether FDA would actually assert that authority over an AI that prescribes on its own, so the answer remains open.
| Who | What it controls | Where clinical AI fits |
|---|---|---|
| States | Clinician licensing and the practice of medicine | Who may prescribe or triage, and which safety checks apply |
| FDA | Medical devices | Likely jurisdiction when AI acts without human supervision, per STAT |
| CMS | Medicare payment and utilization review | Payment codes and pilots such as WISeR |
Each side has a legitimate claim, and that is what makes the fight hard to resolve with a memo. Companies in the middle have a practical problem. An industry analysis cited in the coverage advised vendors piloting in state sandboxes to plan for FDA device filings, according to AI Weekly. A state pilot may not be a shortcut around federal review. It may be the first step toward one.
Where the administration is pulling in two directions
The Trump administration has sought to reduce regulatory burdens on health-AI developers, and it has created new payment pathways for AI tools. At the same time, its agencies are trying to defuse the conflict with state sandboxes, according to STAT and AI Weekly. Those moves pull in different directions. Lighter rules and new payments encourage deployment, while the sandbox effort acknowledges that deployment is outrunning the federal map.
Payment is the clearest example. As of January 2026, 26 CPT codes were authorized for clinical AI solutions in the Medicare Physician Fee Schedule, according to Manatt Health's policy tracker. Only three of them are permanent Category I codes. Most of the payment infrastructure for clinical AI has not yet been made permanent, and that gives federal agencies a reason to coordinate before tools scale.
Congress is part of the picture too. The draft federal framework described by Manatt would apply a three-year preemption to state laws regulating AI model development, not to use or deployment. That line matters. If it held, states would keep control over how clinical AI is used in their own clinics and programs, even as federal law governed how models get built. Meanwhile, the House “Great American AI Act” preemption proposal was still a discussion draft as of late October, according to Questa AI. Federal preemption is not law yet.
The ATA Action proposal and the October 20 comments
The FDA discussion paper on generative-AI medical devices and the response from ATA Action are both dated October 20, 2026. HIT Consultant reported that ATA Action proposed a voluntary federal-state clinical AI sandbox and urged FDA to formalize coordination with CMS. That is a practical ask, and it suggests that at least some health stakeholders see coordination, not a single federal rule, as the fastest fix.
A voluntary sandbox would give states and companies a structured place to test clinical AI while federal agencies watch. It would also make the arrangement depend on who signs up, which means the rules could vary from one program to the next. For patients, that is the trade-off to watch: faster testing, but less certainty about which safety checks apply in each state.
Medicare's AI prior authorization pilot shows the stakes
The CMS WISeR model already puts AI into a high-stakes decision. It requires AI-backed prior authorization for doctors in six states treating some care in traditional Medicare, as Healthcare Dive reported. Prior authorization determines whether a patient gets a service, so errors carry real consequences.
The Electronic Frontier Foundation sued CMS over the pilot to obtain records on vendors and accuracy testing. That suit predates this week's STAT report, so treat it as background rather than a new development. Still, it shows what people want to know when AI shapes coverage decisions: who built the tool and how well it was tested. Those same questions will shape any federal-state sandbox.
States keep legislating while Washington debates
New York's governor signed a youth AI chatbot safety bill on October 28, according to IAPP and the Times Union. USA 24 has already covered that law's age-check provisions, so this piece focuses on its place in the broader fight. The signing shows that states keep acting on AI even while the federal boundary remains unsettled.
States are also using enforcement. Texas sued OpenAI and Sam Altman over ChatGPT safety for minors. Federal agencies are active as well, and the FTC has demanded that Anthropic executives testify in an AI consumer-harm probe.
What patients and clinicians should do now
Readers do not need to take a formal step. A few questions can still help you understand the oversight behind an AI tool used in your care:
- Ask whether a licensed clinician reviews AI recommendations before they affect your treatment or prescription.
- Ask which agency oversees the tool, whether that is FDA, your state medical or pharmacy board, or CMS through a payment program.
- If an AI tool screens or triages your request, ask what safety checks apply and who you can contact with a concern.
Clinicians face a different risk. They remain legally responsible for over-relying on AI outputs, according to blueBriX's analysis of healthcare policy. Using an AI tool does not hand the duty of care to the software, so clinicians working with sandboxed tools should document how they reviewed each recommendation.
What to watch next
Three things will show where this goes. First, whether FDA publicly explains how it will treat an AI that acts without a clinician in the loop, since the sources say that question is still undecided. Second, whether CMS and FDA respond to ATA Action's call for formal coordination. Third, whether a federal preemption bill advances with the split between model development and deployment intact.
Watch which states name their sandbox participants, too. The reporting I could access does not say which states run clinical-AI sandboxes or which tools are inside them. That disclosure would show whether these programs stay narrow or spread quickly. If federal agencies answer before the first large state program scales, patients may get clear rules. If they do not, the first real test of the jurisdiction question will likely come from a tool already in a clinic.
