Monday, November 16, 2026

CMS Publishes Template for Uses of SUPER INTELLIGENCE in Medicare, Medicaid

Recall that recently, HHS ordered that HHS, CMS, NIH, FDA, all use "SUPER INTELLIGENCE" rather than artificial intelligence.  My blog.

We've now got a brand new 8 page report on how cMS wants to use SUPER INTELLIGENCE to lower costs while giving better answers and faster care.  And it's getting some press.

Find the 8 page white paper here:

https://www.cms.gov/files/document/clinical-si-roadmap.pdf

Coverage at Becker Health IT here.  At Orange Slices here.

##
AI CORNER

##


CMS’s AI Roadmap: Better Diagnostics, More Physician Capacity—and Lower Costs

CMS has released an eight-page Clinical Super Intelligence Roadmap, laying out an ambitious federal agenda for clinical AI. For readers in radiology, pathology, and genomics, it deserves attention for both its enthusiasm about technology and its pointed message about reimbursement: CMS wants AI to improve care while bringing costs down.

Read the CMS announcement and the eight-page roadmap. The document was posted November 13.

The roadmap’s subtitle is Better Answers, Faster Care, and Lower Costs. Those objectives could apply directly to software that helps a radiologist complete an interpretation, a pathologist sign out a case, or an oncologist understand a complex molecular report. Much of the immediate policy discussion concerns primary care and chronic disease, but the implications for diagnostic medicine are substantial.

The payment message deserves particular attention. On page 7, CMS states:

“Technology should be deflationary to costs.”

The surrounding discussion favors payment arrangements that reward clinical outcomes and demonstrated value, and warns against incentives that increase total spending. Developers looking for a succession of new, separately payable AI services should read this passage carefully. CMS is interested in financing better care, but expects the economics of software to produce affordability.

This aligns closely with Chris Klomp’s recent public remarks, covered in my November 8 discussion of his Senate hearings and Wall Street Journal interview. In the Senate, Klomp emphasized aligned incentives, appropriate preventive spending, and Medicare’s independent coverage judgment. His subsequent interview made the AI financing argument more explicit: reimbursement should translate clinical productivity into lower costs, with ACCESS as an example. The new roadmap gives that philosophy a broader policy expression.

Improve and Efficient

For diagnostics companies, my reading is that CMS will increasingly ask two questions: Does this technology improve care? And how do its efficiencies benefit patients and purchasers?

An algorithm might save physician time and still increase Medicare spending if every use generates an additional payment while all existing payments continue. Conversely, a practice might purchase useful software from its operating budget because it increases capacity, improves turnaround time, or makes work easier. An organization accountable for total spending might invest because better diagnostic decisions reduce downstream costs. These are different commercial models, and the roadmap’s payment philosophy makes those differences important.

CMS organizes its agenda around four priorities: public trust, usable health data, clearer regulatory pathways, and payment for demonstrated value. It also contemplates greater autonomy as evidence accumulates, including future AI-led accountable care arrangements. These are policy ambitions; the document does not establish new Medicare coverage or payment rates for diagnostic AI.

Better than Today

One useful phrase is “better than today.” CMS argues that AI should be assessed against a health system already affected by delayed access, fragmented records, and preventable errors. Applied to diagnostics, that suggests evaluating the complete clinical workflow: accuracy, physician time, report quality, turnaround time, and the consequences for patient management.

[Note - could insert here, recent review papers that few AI papers show "improvement" or "clinical utility."]


AI-generated reports provide a particularly relevant example.

CMS’s separate Technology-Enabled Care and Artificial Intelligence explainer discusses AI scribes, diagnostic assistance, and machine learning that complements imaging review. Those functions represent different levels of clinical involvement. Recording a physician’s words is one task; analyzing an image and proposing findings is another.

A concrete example is See-Mode’s SMART-B, cleared by FDA in July 2026. Its public 510(k) summary describes breast ultrasound software that localizes and characterizes lesions, generates BI-RADS descriptors and a category, and produces a report for physician review and approval. The physician can modify the report before finalization. Here, image analysis connects directly to report production.

That connection could become commercially decisive. Software that helps finish a case can address a larger portion of the physician’s work than a tool that supplies one additional alert.

  • Consider a hypothetical system that reduces total physician time per case by 40%, including review and corrections. With unchanged case mix and no other bottlenecks, the same physician time could support approximately 67% more cases. 
  • That is illustrative arithmetic, not a performance claim about SMART-B. It shows why substantial, reproducible time savings could produce extraordinary physician enthusiasm and purchasing interest.

The important measure is time to a reliable, signed report. A rapidly generated draft that requires extensive checking may deliver little benefit. Human factors, interface design, and integration with existing systems become central to clinical and commercial success.

Pathology has a parallel opportunity. A future system might combine slide analysis, proposed diagnostic language, microscopic descriptions, structured reporting fields, and checks for missing information. Each capability would need evidence appropriate to its clinical function and intended use. The potential attraction is a coherent working environment that helps the pathologist complete the case.

Computational pathology already has regulatory precedents for other functions. FDA granted De Novo authorization to ArteraAI Prostate in July 2025, classifying it as software analyzing digital images for prostate cancer prognosis. Its FDA record also identifies an authorized predetermined change control plan. That is a different application from report drafting, but it illustrates the clinical value that can be extracted from tissue images and a mechanism for managing specified software changes.

Acting on Genomic Results

For genomics, the most relevant opportunity may lie between generating a result and acting on it. Molecular findings need to be interpreted alongside histology, imaging, prior treatment, and current clinical evidence. A plausible application would assemble those elements into a traceable draft interpretation for specialist review. Another could organize serial molecular residual disease results alongside imaging and treatment history.

These are potential applications of the roadmap’s emphasis on information integration. Their value would depend on whether they improve decisions and outcomes. AI interpretation would not, by itself, establish the clinical utility of a genomic test or a treatment strategy.

The existing CMS ACCESS model and FDA TEMPO pilot illustrate coordination between payment and evidence generation. Their current scope concerns specified chronic conditions, rather than a general pathway for oncology diagnostics or AI reporting. Nevertheless, they provide an example of agencies linking technology deployment to measurable results.

The roadmap also discusses possible competency-based evaluation of certain AI devices, continuing performance surveillance, and predetermined change control plans. For diagnostic developers, the practical questions will include performance across institutions, scanners, specimens, patient populations, and software versions.

BOTTOM LINES?

My strongest takeaway is that clinical productivity could become a powerful adoption trigger in radiology and pathology. Software that substantially reduces the effort required to produce a reliable report could attract physicians and institutions even before a dedicated reimbursement pathway emerges.

But Klomp’s remarks and the new roadmap add a second requirement to that commercial opportunity: CMS expects technological productivity to help restrain spending. Companies will need to explain both why clinicians want the product and how its value reaches the people paying for care.

And track how this aligns with CMS reimbursement efforts like SaMD, and AMA CPT with "Appendix S" and "CMAA" and more to come.