AI prescribing (AI prescriptions)
AI prescribing — often searched as “AI prescriptions” or “prescribe using AI” — describes the use of artificial intelligence in the prescribing process; in the United States AI cannot legally prescribe medication, so in practice the term means AI that prepares, checks, and routes prescription requests for a licensed clinician to decide and sign.
- No state grants prescriptive authority to software
- A licensed clinician’s NPI is on every prescription
- Utah’s sandbox pilot: AI renewals only, with physician review
- FDA exempts reviewable clinical decision support from device rules
Can AI prescribe medication?
No. Prescriptive authority is conferred by state licensure on physicians, nurse practitioners, physician assistants, and a small set of other professions, each within a defined scope. No state confers it on software, and every e-prescription must identify a licensed prescriber by NPI. The edge is being tested in one place: Utah’s AI regulatory sandbox has granted one company temporary relief so its AI can renew existing prescriptions for roughly 200 maintenance medications — renewals only, no controlled substances or injectables, physician review in the current phase — without changing who may prescribe. A federal bill introduced in 2025 would let FDA-authorized AI prescribe where a state permits it; at the time of writing it remains in committee.
So when a product says it can “prescribe using AI,” one of two things is true: a licensed clinician is deciding and signing behind the interface, or the product is out of bounds. The first is a legitimate and increasingly common design. The second is a licensing problem.
What AI actually does in prescribing today
Nearly everything except the decision. The work that used to make prescribing slow — gathering a history, checking it against a protocol, screening interactions, assembling the request, chasing the refill — is exactly the work AI handles well.
- Structuring intake so the clinician sees a complete, comparable case
- Screening against protocol criteria, contraindications, and interactions before review
- Drafting a proposed plan — “continue, step dose per protocol” — for the clinician to accept or change
- Checking refill eligibility, prompting a PDMP query, and queueing the request
- Carrying out the signed decision: routing to pharmacy, tracking, patient updates
The regulatory frame
Three bodies of rules bound AI in prescribing. State practice acts define who may prescribe and, for telehealth, which modalities are lawful for which drugs. FDA regulates software that makes clinical decisions as a medical device, while exempting clinical decision support that a clinician can independently review. And a growing set of state AI laws — California now requires disclosure when generative AI communicates clinical information to patients, and requires licensed-physician review of AI-driven coverage decisions — add disclosure and human-review duties on top.
None of these prohibit AI in the workflow. All of them assume a licensed human owns the decision. Build to that and the rules are on your side.
Building with AI in prescribing without crossing the line
Treat every AI output as a draft for a licensed clinician; never let the model’s output become an order without a human action. Keep prescribing inside infrastructure that enforces licensure and state modality rules regardless of what the AI proposes. Disclose AI involvement to patients where state law requires it. Log what the model saw, what it proposed, and what the clinician decided. And keep controlled substances on a stricter, separate protocol.
Compliance handled, so you can build
Lithos runs the clinicians, pharmacies, and 50-state rules behind your care program — one API.
Frequently asked questions
Can ChatGPT or another AI assistant prescribe medication?
No. An assistant can help a user request care and can call infrastructure that routes the request to a licensed clinician, who decides and signs. The assistant itself has no prescriptive authority in any state.
Is it legal to use AI to write prescriptions?
Using AI to draft, screen, and assemble a prescription for a licensed clinician to review and sign is legal and common. Letting AI issue a prescription without a clinician’s decision is not.
Is Utah allowing AI to prescribe?
Utah is running a supervised sandbox pilot, not a change in law: for a one-year term the state is not enforcing unlicensed-practice rules against one company whose AI renews existing prescriptions for roughly 200 maintenance medications. New prescriptions, dose changes, controlled substances, and injectables are excluded, and a physician currently reviews each renewal.
Will AI be allowed to prescribe in the future?
The edges are moving — Utah’s sandbox pilot, similar sandboxes forming in other states, and a 2025 federal bill that would allow FDA-authorized AI to prescribe where states permit. None has changed who may prescribe at the time of writing. Programs should build for today’s rules and treat any change as upside.
What is the difference between AI prescribing and e-prescribing?
E-prescribing is how a signed prescription travels to the pharmacy electronically. AI prescribing describes AI helping prepare the request before a clinician signs. In a modern program both are present: AI assembles, a clinician signs, e-prescribing transmits.