Article
Nov 4, 2026
What Cassidy's PATIENT AI Act would change for clinical AI
A Senate draft bill would change how clinical AI is reviewed, certified and paid for by Medicare. Here is what it does, what it leaves open, and why it matters.
The chair of the Senate health committee has released a draft bill that would rewrite how AI tools in American health care are reviewed, certified and paid for. It is only a discussion draft, and its sponsor leaves office in January, but it lays out a detailed federal blueprint for clinical AI, and comments are open until December 1.
Here is what is in it, what is missing, and why clinics on both sides of the border should pay attention.
What happened
On October 29, 2026, Sen. Bill Cassidy, the Louisiana Republican who chairs the Senate Health, Education, Labor and Pensions (HELP) Committee, released a discussion draft of the PATIENT AI Act, short for the Protecting Americans Through Intelligent, Ethical, and Transparent Artificial Intelligence in Health Care Act. The committee posted the full draft text and a section-by-section summary, and asked stakeholders to send feedback to patientai@help.senate.gov by December 1.
Cassidy said implementing AI in health care "must be done responsibly and to the benefit of patients, innovators, and consumers." The committee framed the goal as "creating regulatory predictability while prioritizing patient safety."
Who would regulate which AI tools
Today, many clinicians are unsure whether a given AI product is a regulated medical device or simply software. The draft tries to settle that by splitting oversight between two agencies:
The FDA would keep higher-risk tools and get a risk-based review process for software-based devices, including authority to accept planned software changes before they ship and review updates on a rolling basis.
The Office of the National Coordinator for Health IT (ONC) would certify AI applications that are not FDA-regulated devices, by adding them to the definition of certified health IT.
Third-party certification bodies, approved by HHS, could review low- to moderate-risk software.
The practical effect would be faster updates for AI products already on the market, and a formal certification route for tools such as documentation and administrative software that sit outside FDA device rules.
A Medicare payment pathway for AI
Five of the draft's sections deal with the Centers for Medicare and Medicaid Services. They would direct HHS to build a framework for deciding coverage of AI-based services, create a dedicated coverage evaluation pathway, propose payment methods, and set up a process for billing AI services with HCPCS Level II codes.
The most striking provision asks the CMS Innovation Center to test a payment model for AI applications "capable of performing a clinical function autonomously." According to Medical Daily's reading of the draft, the model would run for five years and must start no later than 24 months after the law is enacted. Medicare also could not decide a service lacks clinical value "solely because" the AI works without a physician interpreting results at the same time.
Participants would have to give patients "informed beneficiary notice" that AI is part of their care and maintain "appropriate clinical accountability and escalation arrangements." The draft does not define "autonomously," set payment rates, or list which tasks or conditions would qualify.
What the draft leaves open
Liability
The question most clinicians ask first, who is responsible when AI gets it wrong, is not answered. The draft directs HHS to work with a nonprofit to study liability and publish recommendations. Separately, it states that existing federal preemption of state medical device product liability laws applies to software-based devices, which would mainly shield manufacturers of FDA-cleared products.
Privacy and transparency
HHS would issue guidance on meeting HIPAA privacy and security rules when covered entities use AI. The Government Accountability Office would review proposals for transparency and explainability and report back. Both are studies and guidance, not new rules.
Its political future
Cassidy lost his Republican primary in May, and his term ends in early January, so a formal bill would likely need a new sponsor or reintroduction, as Medical Daily notes. Drafts like this often shape later legislation even when they do not pass themselves.
How it fits with other proposals
The draft is not the only federal move on clinical AI. The House Doctors Not AI Act, introduced September 1 by Reps. Greg Landsman, Buddy Carter, Kim Schrier and Tom Barrett, would require that medical necessity denials come from a licensed clinician and that insurers disclose AI involvement. Carter said AI "should never have the final say in denying someone access to medically necessary care."
Not everyone is ready for autonomous care. American Medical Association CEO Dr. John Whyte told The New York Times that models built to independently provide care are "not ready for widespread use," as reported by Medical Daily.
What this means for clinics
Expect more formal certification. If ONC certification expands to non-device AI, clinics may soon be able to ask vendors a simple question: is this tool certified, and by whom?
Autonomous AI is the exception, not the rule. The payment model is a test. For now, most clinical AI is designed to support a clinician, and Medicare billing still assumes one.
Keep a human in the loop and document it. The draft's emphasis on clinical accountability and escalation is a good standard for any clinic today. Know who reviews AI output, and when.
Canadian clinics should watch too. This draft does not change Canadian law, but many AI products sold in Canada are built for the US market first, and US rules shape how those products are designed and documented.
You can comment. Clinicians and clinic owners can send feedback to the committee until December 1.
At Athel, we build around that accountability model: our AI co-clinician interviews patients before the visit by voice, web chat or video and gives the physician a structured history, and the clinician reviews every summary before it is used.