20 November 2026 · Medows
Clinical AI's New Federal Bar: Better Than Today
A federal roadmap holds clinical AI to a "better than today" bar, and admits the EHR moved work onto screens without redesigning workflow.
On November 13, HHS and CMS published "A Clinical Super Intelligence Roadmap: Better Answers, Faster Care, and Lower Costs." It is a federal framework for pushing clinical AI into American medicine faster: payment, regulation, data, trust.
Most of it reads like policy. One sentence reads like it was written by someone who has used a hospital computer at 2 a.m.
The line that lands
"The internet made information available. The EHR digitized the chart. But neither one consistently reduced the cognitive and administrative burden on patients or clinicians. In some cases, the EHR moved work from paper to screens without redesigning the workflow, making it more complicated for clinicians to deliver care."
That is the United States government writing down the thing in a published document. Digitising the chart is not redesigning the shift. Two decades of clinical software assumed those were the same project. They are not, and the people who found out were the ones holding the trolley key.
"Better than today" is an honest bar
The roadmap's first guiding principle is to hold clinical AI to a "better than today" standard for safety, effectiveness, access, and outcomes, "while recognizing that the current system is already imperfect."
A later section titled "What Will Not Be Done" is blunter. The roadmap "does not hold SI to a standard of perfection: it will not wait for flawless evidence to open pathways, but it will require evidence to keep them open."
Read that second clause again, because it is the load-bearing half. Evidence to keep a pathway open means somebody has to measure the tool after it ships, in use, on real patients. The roadmap commits to exactly that: real-world post-market surveillance, plus predetermined change control plans so a model can improve without quietly becoming a different model.
That is a reasonable trade. It also moves the burden of proof from a regulator's desk to the ward.
Whose "today" is the baseline
Here is where the document shows its angle. The evidence it cites for AI beating the current baseline is this:
- OpenAI's o3 ranked the final diagnosis first in 60% of complex cases, against 15.9% for internal medicine physicians.
- Microsoft's MAI-DxO correctly diagnosed 85.5% of challenging cases, against roughly 20% for practising physicians.
- Cedars-Sinai's system gave optimal treatment recommendations in 77.1% of real patient cases, against 67.1% by physicians.
Complex cases. Challenging cases. Recommendation comparisons. Good results, properly cited. But notice what the human comparator is doing in every one: reading a case. Not carrying twenty-eight patients, not being interrupted mid-sentence, not deciding at hour eleven which of four unwell people to see first.
The roadmap's own definition of success is "better access and affordability for patients, more leverage and sustainable adoption for providers, and appropriate rewards for innovators." Patients get access. Innovators get rewards. The clinician's day appears as "leverage."
The trust number the roadmap half-quotes
To make the case that public resistance will fade, the roadmap cites a Pew Research Center survey: 60% of U.S. adults would be uncomfortable with their provider relying on AI for things like diagnosis and treatment. It then notes that "sentiment can shift quickly."
Maybe. The same Pew report, fielded among 11,004 adults from December 12 to 18, 2022, also found that three-quarters of Americans were more worried that providers would move too fast on AI before understanding the risks than too slow. The roadmap quotes the discomfort figure and argues for speed. Both findings come from the same survey.
The document is fairer about doctors. It says providers "want clarity on liability, evidence, and workflow before they lean in." Three things. A roadmap can deliver liability clarity and it can fund evidence. Workflow is the one no federal document can legislate. That gets built or it does not.
What changes on Monday
Nothing. The roadmap does not mandate adoption and does not create a new agency. It is direction, not duty.
But direction compounds. Payment is being steered toward outcomes, regulation toward monitoring after deployment rather than only before it. If the evidence that keeps a pathway open has to come from real use, then the ward becomes the instrument, and the record of what the model suggested, what the clinician did instead, and why, becomes the evidence base.
That is a documentation problem before it is a policy problem. Which is the part we keep coming back to: an AI suggestion is only worth as much as the trail it leaves behind it.
Sources
- A Clinical Super Intelligence Roadmap: Better Answers, Faster Care, and Lower Costs (CMS, November 2026)
- Trump administration releases roadmap to boost clinical AI (Nextgov/FCW, November 16, 2026)
- 60% of Americans Would Be Uncomfortable With Provider Relying on AI in Their Own Health Care (Pew Research Center, February 2023)
Medows is a clinical AI workspace for the doctor on rounds. Learn more or write to alapan@medows.ai / alapanx@gmail.com.