How are AI doctors regulated in the US?

Does the FDA regulate AI doctors?

The FDA regulates medical devices, and software can be one. Whether a given AI tool is a device depends on what it is meant to do and who it is meant for.

The 21st Century Cures Act, enacted in December 2016, added section 520(o) to the Federal Food, Drug, and Cosmetic Act, which takes some software out of the device definition. One carve-out covers clinical decision support (CDS) software for health care professionals. To be left out, a CDS tool has to meet all 4 criteria in the law.

The FDA's CDS guidance, issued January 29, 2026, is direct about decision support aimed at the public: software that supports or gives recommendations to patients or caregivers, not health care professionals, meets the definition of a device. The FDA says its existing digital health policies still apply to those tools, including policies under which it does not intend, for now, to enforce device rules on some lower-risk software. FDA guidance is nonbinding: it shows the agency's current thinking, not a new law.

• Criterion 1: it is not meant to acquire, process or analyze a medical image, a signal from a lab (in vitro diagnostic) test, or a pattern or signal from a signal-collecting device.
• Criterion 2: it displays, analyzes or prints medical information, like a patient's record or clinical guidelines.
• Criterion 3: it supports or gives recommendations to a health care professional.
• Criterion 4: it lets that professional review the basis for the recommendation, so they don't rely mainly on the software to decide.

Has the FDA cleared any AI medical devices?

Yes. The FDA keeps a public list of AI-enabled medical devices that have been authorized for marketing in the US after a review of safety and effectiveness. Each entry shows the company, the device, and the date of the FDA decision.

The FDA says the list is not complete, because it finds devices mostly by AI-related terms in their summaries. It also says it plans to look for ways to tag devices that use large language models in a future update. So the list is a useful starting point, but a product missing from it doesn't prove anything on its own.

Can an AI hold a medical license?

Not today. Medical licenses are issued to clinicians by state boards. California's law, for example, says anyone who diagnoses, treats or prescribes for a condition without a valid certificate is guilty of a public offense, punishable by a fine, jail or both.

So in general, a licensed clinician makes the call on prescriptions and formal diagnoses, even when AI helps. Utah is testing a narrow exception: its Office of Artificial Intelligence Policy runs a temporary pilot that lets a company's AI tool renew some existing prescriptions first written by a licensed physician. The AI can't write new prescriptions, handle controlled or addictive drugs, or change treatment plans, and in the first phase, which the state says the pilot is still in, a licensed physician reviews each renewal before it goes to a pharmacist. Utah says these agreements are temporary and conditional, and none of this gives the AI a license.

Our page on whether AI doctors can prescribe covers that side in more detail.

What state laws cover AI in health care?

States are starting to pass laws that name AI directly. California is one example. The details below come from its Senate Judiciary Committee bill analysis and its Medical Board.

Other states have proposed or passed their own rules. They change quickly, so check your state legislature's website for the current text.

• California AB 489 (Chapter 615): stops developers and deployers of AI from using terms, letters or phrases, in ads or in the product itself, that imply the AI holds a health care license. It also bans implying that care or advice comes from a licensed person when it doesn't. Each use counts as a separate violation, and the matching state licensing board can enforce it, including by asking a court for an injunction.
• California Health and Safety Code 1339.75: when a clinic, health facility or doctor's office uses generative AI to write messages about a patient's clinical information, the message has to say it was AI-generated and explain how to reach a human. This doesn't apply if a licensed provider read and reviewed it first.

What does the FTC do about AI health apps?

The FTC's job is to prevent unfair or deceptive acts or practices, including in health ads. Its health products guidance covers health-related apps along with supplements, tests and health equipment.

The FTC expects health claims to be backed by competent and reliable scientific evidence. For health benefits, that generally means randomized, controlled human trials. Customer stories alone are never enough. The guidance also says the FTC follows the same basic steps for any health claim, whatever category it falls into under FDA law, and that anyone who takes part in deceptive marketing can be held liable.

What would licensing for AI doctors look like?

Pymander has published a working draft bill, the Clinical AI Provider Licensing and Patient Safety Act, written for New Hampshire's 2027 session. It has not been introduced and is not law.

The draft proposes 4 license levels. An AI starts with a licensed clinician making every decision, and moves up only after it shows it is safe. It would give patients the right to a human review of any decision and a full copy of their record. Serious reportable events would go to the state within 72 hours, or within 24 hours if they present ongoing imminent danger. The draft keeps controlled substance prescribing, surgery, involuntary treatment and emergency stabilization out of an AI's independent authority unless another law allows it.