What the draft contains
Senator Bill Cassidy, the Republican who chairs the Senate Health, Education, Labor and Pensions Committee, released the text in late October. Its full name is the Protecting Americans Through Intelligent, Ethical, and Transparent Artificial Intelligence in Health Care Act. No vote has taken place. The committee is collecting written comments until 1 December 2026.
Four provisions carry most of the weight. The draft would give the Food and Drug Administration authority to review changes to AI software on a rolling basis, rather than only at fixed approval points. It splits oversight between the FDA and the Office of the National Coordinator for Health Information Technology. It directs the Department of Health and Human Services to use AI internally and to publish privacy and security guidance. And it opens a route for AI tools to be reimbursed through existing payment channels.
A model that changes every month
Device regulation was built around products that stay the same after clearance. A diagnostic model retrained on new hospital data every few weeks does not fit that pattern. Cassidy's draft responds by letting the FDA look at updates as they ship, instead of forcing vendors into a fresh submission each time the weights move.
The trade-off sits in the same provision. Faster review also means less pre-market evidence than clinicians and patients are used to seeing. The draft pairs the rolling process with what it describes as reinforced oversight for patient safety and privacy. The comment period is where that balance gets argued.
Two agencies, one boundary
Splitting the FDA from the ONC matters more than it sounds. The FDA regulates devices. The ONC sets standards for health IT and electronic records. A chatbot that drafts a discharge summary and a model that flags a suspicious scan sit on different sides of that boundary, and vendors have spent years arguing about which side they are on.
The draft draws a line between them. It does not settle every case, and patient groups and clinicians disagree about where the line belongs.
Congress is moving on several tracks
In September 2026 a bipartisan group in the House introduced the Doctors Not AI Act, which would keep medical necessity and claims determinations with licensed clinicians. The Pennsylvania House passed HB 1925 on 30 September 2026, requiring transparency when AI is used in clinical or insurance decisions. In October the Senate passed Cassidy's separate Health Care Cybersecurity and Resiliency Act by unanimous consent. The PATIENT AI Act is the piece that deals with approval and payment.
What Europe does differently
There is no European equivalent of this bill. Medical software that qualifies as a device is certified under the Medical Device Regulation and carries a CE mark. The EU AI Act then adds obligations on top, because a safety component of a medical device counts as high-risk.
The calendar moved. Under the Digital Omnibus, Regulation (EU) 2026/1744, the compliance deadline for standalone Annex III high-risk systems was postponed to 2 December 2027, and for systems embedded in products to 2 August 2028. A hospital in Prague or Lyon buying an AI triage tool this month is working with a rulebook whose high-risk chapter is not yet fully in force.
GDPR sits underneath all of it. Health data is a special category under Article 9, so a clinic needs a legal basis before sending patient records to a model provider, whether the model runs in Frankfurt or on a server in Ohio.
One part of the EU framework is already live. Since 2 August 2026, Article 50 transparency obligations require deployers to tell people when they are interacting with an AI system, and require synthetic audio, image and video to be labelled and watermarked. A practice using an AI receptionist or an AI-written patient leaflet is covered now, not in 2027.
National authorities and the EU AI Office also enforce the general-purpose AI rules directly, with documentation, systemic risk mitigation and copyright compliance required from providers whose models are on the European market. The Commission publishes the underlying texts at digital-strategy.ec.europa.eu.
Liability is the question doctors ask first
One third of US physicians say a clear liability framework is the single main factor in whether they adopt clinical AI, according to the American Medical Association. The Pew numbers point the same way from the patient side: 41% want AI to play no role in diagnosis at all, and 46% reject AI involvement in insurance claims.
Translation is the exception. About 80% support AI handling translation, with 26% comfortable leaving it fully automated and 52% wanting human review. Roughly one in twenty is happy to hand over diagnosis.
Those figures help explain why the draft spends as much space on payment and process as on technology. Reimbursement decides whether a tool reaches a clinic. Liability decides whether a clinician switches it on.
The deadline for comments
The consultation closes on 1 December 2026. After that the text can change substantially before it becomes a bill, and it would still need to pass both chambers. For now the document is a draft with a comment window, and the Pew survey is a snapshot of US opinion rather than a European one. European rules continue on their own schedule, with the embedded high-risk date set at 2 August 2028.
Does the EU AI Act require approval before a medical AI tool can be sold in Europe?
No. The AI Act is not a market authorisation regime for medical software. A tool that qualifies as a medical device is certified under the Medical Device Regulation and gets a CE mark; the AI Act adds requirements for high-risk systems on top of that. The FDA model in the US draft, with the agency reviewing updates as they ship, has no direct EU counterpart.
Would the PATIENT AI Act affect European vendors selling into the US?
Any company placing a medical AI product on the US market already works under FDA device rules, whatever its headquarters. The draft would change how updates to AI software are reviewed, and the committee is accepting comments on the text until 1 December 2026.
Are the Pew figures about Europe too?
No. The survey was fielded in June 2026 and published by Pew on 29 October 2026; it covered 3,488 US adults and measured American attitudes. The article uses it to explain the political pressure behind the Senate draft, not as a description of European opinion.
