Essay by Chat GPT. See short blog pointing to RFK video (August 26,, 2026).

###

CPT’s “Social Bias” Requirements Land in a Very Different Washington

A small addition to the American Medical Association’s CPT application process illustrates how quickly the political meaning of technical language can change.

Beginning with applications reviewed at the September 2024 CPT Editorial Panel meeting, the AMA added a series of questions for proposed medical services involving artificial intelligence or software. Two of the questions are especially notable:

Generalizability
“Please describe efforts to ensure broadest generalizability of this software, e.g. curation of training databases, plans for surveillance of real-world data, etc., (give references).”

And immediately afterward:

Social Bias
“Characterize the potential for perpetuation, propagation, or mitigation of social bias, which might reasonably be anticipated (give references).”

These were genuinely new questions. Barbara Levy, then vice chair of the CPT Editorial Panel, explained in a November 2024 Cancer Letter article that the panel had been receiving an influx of AI applications that were sometimes unclear about what the algorithm actually did. She said the additional questions were intended to help CPT understand AI-based services more precisely. The article reproduced all eight new AI questions, including the two above. (Open access at the journalist's blog; McKenzie Prillaman)

That history is important, because it shows that the language was not written for the political environment of 2026. It was written in 2024, near the end of the Biden administration, when “health equity,” “algorithmic fairness,” “bias mitigation,” “representative datasets,” and related terminology were deeply embedded and often required in federal and academic discussions of healthcare.

Two years later, those same words have landed in a very different Washington.

Generalizability Is Not a Special DEI Problem

There is a scientific case for AMA’s first question.

A medical algorithm developed at three academic medical centers may not perform equally well when deployed in community hospitals. A dermatology algorithm trained primarily on light skin can fail on darker skin. A pathology algorithm may be sensitive to staining protocols or scanner platforms.

Those are conventional questions of external validity, transportability, analytical performance and patient safety. Asking an applicant about training populations, relevant subgroups and real-world surveillance isn't unreasonable.

The intellectual lineage is easy to see.

A 2023 open-access npj Digital Medicine article, “Considerations for addressing bias in artificial intelligence for health equity,” discussed how bias can enter healthcare AI, how it can affect different populations, and how it can be identified and mitigated across an AI product’s lifecycle. Several authors were senior FDA Center for Devices and Radiological Health officials. Another author, Michael Abràmoff, was a member of the AMA Digital Medicine Payment Advisory Group’s AI Workgroup. (PubMed)
Abràmoff had also coauthored the 2022 npj Digital Medicine paper explaining development of CPT Appendix S, the AMA taxonomy distinguishing assistive, augmentative and autonomous AI. That paper explicitly arose from an AMA workgroup and was intended to help innovators construct CPT code-change applications for machine-performed medical services. (PubMed)

There is no evidence that the AMA copied its 2024 CPT application text, verbatim from either paper. But the conceptual genealogy is not hard to reconstruct. And language in the 2022, 2023, and 2024 documents align.

“Social Bias” Is Something Different

The problem begins when CPT moves from asking whether services work properly across the intended patient population to asking whether they will cause the “perpetuation, propagation, or mitigation of social bias.”

That is not quite the same inquiry.

Suppose a company develops an AI-based genomic test for cancer. It is completely reasonable to ask:

Does the test demonstrate comparable clinically relevant performance in the populations for whom it is intended?

That is answerable with data.

It is considerably less clear what the company is supposed to do with:

Characterize the potential for perpetuation or propagation of social bias.

Where are physicians, or most medical scientists, trained in the sociology of whether new technologies will 'propogate social bias?" Where are the AMA's instructions for these terms and expected research ("add references.")

What counts as “social bias”? How indirect will an AMA reviewer expect the causal chain be? Where are his/her instructions?

Is the applicant expected to discuss unequal access to sophisticated oncology centers? Or the tragedy of historical racial disparities? Or will the Administration view those answers as "un-American?"

Insurance coverage? Socioeconomic differences in genetic testing? The possibility that genomic classifications themselves could reinforce stereotypes? Structural inequities several steps removed from the actual analytical performance of the test?

AMA deliberately provided no guardrails or stopping rules.

That is why the historical contrast with ordinary CPT coding is striking. When PET imaging was developed, CPT did not ask whether three-dimensional reconstruction of positron emissions might perpetuate racial inequities. When robotic surgery emerged, code applicants were not asked whether access to surgical robots might propagate socioeconomic harms. When new pathology procedures were coded, the applicant was not normally required to characterize whether the technology might reinforce structurally embedded social disparities or racism.

There were plenty of scientific questions. There were questions about clinical use, efficacy, distinctiveness of the service, FDA status where applicable, adoption, and eventually valuation.

And It Really Is a Required Question

There is an additional point that makes the wording more consequential than an aspirational AMA ethics statement.

The AMA’s current CPT application FAQ explains how staff determine whether an application is complete. AMA staff “verifies that all application questions have been answered.” If they have not, the application is returned to the applicant for completion within a defined time period. (American Medical Association)

Thus, for an application to which these AI questions apply, this is not merely optional reading material about responsible innovation. The company cannot simply decide that a discussion of “social bias” falls outside the purpose of a coding application and leave the box blank.

It must answer.

A Linguistic Fossil From a Different Administration

The wording is especially interesting because it now reads almost like a linguistic fossil from another policy era.

Indeed, there is something slightly séance-like about it. In 2026, CPT appears to be channeling the language of peak 2022–2024 Biden-era health-equity and responsible-AI policy into an administration that has explicitly repudiated much of that vocabulary.

This is not an inference based merely on campaign rhetoric.

President Trump’s July 2025 executive order “Preventing Woke AI in the Federal Government” identifies DEI as an ideology that can distort AI and expressly includes within its description of DEI concepts such as “unconscious bias,” “intersectionality,” and “systemic racism.” The order establishes “truth-seeking” and “ideological neutrality” principles for large language models procured by the federal government. (The White House)

The administration has used still stronger language elsewhere.

A January 2025 executive order on education describes certain equity-oriented teachings as “radical, anti-American ideologies” and specifically objects to requirements involving “White Privilege” and “unconscious bias.” (The White House)
A March 2025 White House order and fact sheet concerning the Smithsonian and federal historical sites directs the administration to eliminate “improper, divisive, or anti-American ideology” and attacks what it describes as race-centered narratives portraying American institutions as fundamentally oppressive. It specifically criticizes programs discussing institutional racism and racial systems of power and directs Interior to review historical markers and other materials that “inappropriately disparage Americans.” (The White House)
By July 2026 the White House had gone even further, issuing another executive order after a review of the Smithsonian. The administration ordered temporary National Park Service signage outside the National Museum of American History warning visitors about what it called the museum’s ideological capture. (The White House)
The White House’s underlying report is unusually explicit: in criticizing Smithsonian educational programming, it juxtaposes discussion of “systemic racism” with what it calls an “anti-American ideology.” (The White House)

The administration has repeatedly placed concepts such as systemic racism, unconscious bias and race-centered structural analysis within a category of DEI or ideological frameworks it considers objectionable, and in some contexts it has quite literally labeled related narratives anti-American ideology.

Then Came RFK Jr. — Directly at CPT

AMA CPT is already under extraordinary scrutiny from the same administration.

In July 2026, CMS placed a Request for Information inside the proposed 2027 Medicare Physician Fee Schedule examining the AMA’s CPT system and the RUC. The language is unusually confrontational for a Medicare rule. Among other things, CMS asks for evidence concerning harms associated with “AMA’s monopoly over CPT-4 licenses” and cites longstanding concerns about federal reliance on a private organization with what the rule calls an “obvious conflict of interest.” (MSV)

On August 27, 2026, HHS Secretary Robert F. Kennedy Jr. personally released a roughly two-and-a-half-minute video urging Americans to submit comments on AMA control of CPT.

This is worth emphasizing.

The Secretary of Health and Human Services did not leave the CPT RFI buried in a 1,500-page Medicare regulation for coding specialists and Washington lawyers to discover. He took the issue directly to the public.

Medical Economics reported that Kennedy described AMA’s CPT position as a “de facto monopoly,” criticized licensing charges, and explicitly directed viewers to the Medicare comment docket. The article also carefully notes the limits of the RFI: CMS has not actually proposed how to replace the CPT, and some of Kennedy’s numerical assertions didn't include supporting documentation. (Medical Economics)

Still, the sitting HHS Secretary has personally decided that AMA control of CPT is a top priority issue worth taking to the American public on video.

That is a dramatically different situation from a technical disagreement between CMS career staff and AMA coding experts.

And it makes the new “Social Bias” language much more combustible.

Other Pressures Are Already Present

The July RFI is not AMA’s only CPT problem.

On August 13, PatientRightsAdvocate.org filed suit against AMA in federal court in Chicago. The group seeks a declaration permitting it to republish CPT freely and challenges AMA’s copyright position. The litigation remains at an early stage; there has been no ruling on the merits. (Justia Dockets & Filings)

Again, the “social bias” question has essentially nothing to do with copyright law.

But the institutional backdrop matters. AMA is simultaneously defending a coding system whose government-backed importance and licensing structure are under attack in court, facing a CMS RFI that explicitly uses the word “monopoly,” and being criticized personally by the HHS Secretary.

That is a very inopportune moment to discover that the CPT application form contains a mandatory section headed:

SOCIAL BIAS

followed by:

“Characterize the potential for perpetuation, propagation, or mitigation of social bias.”

One does not need to speculate very far about what an administration official hostile to DEI might make of it.

The Unforced Error

The most important point, therefore, is not that AMA suddenly invented an absurd problem in 2024.

The language arose from a serious and recognizable medical-AI literature. In the Biden-era policy environment in which it was written, the words “health equity,” “bias mitigation,” “representative populations,” and “social bias” occurred together so often that the CPT question probably seemed like a reflex to the people writing it.

The underlying science has not changed. The political meaning of the vocabulary has.

In 2024, asking an AI developer to discuss the “perpetuation, propagation, or mitigation of social bias” sounded like responsible-AI housekeeping.

In 2026, the federal executive branch has formally attacked DEI, expressly identified “systemic racism” and “unconscious bias” as part of the ideology it wants removed from federal AI, labeled certain race-centered narratives “anti-American,” ordered changes at the Smithsonian and national historical sites, and—most importantly for AMA—seen its own HHS Secretary go directly before the public to challenge AMA’s control of CPT.

And this is not merely an AMA position paper that an applicant is free to ignore. AMA’s own application rules say that all application questions must be answered before an application is complete. (American Medical Association)

That is where a technical wording problem becomes an institutional one.

A medical-technology company should certainly be prepared to show that its product works across the patients for whom it is intended. An AI company should be able to explain its training data, external validation, subgroup performance and real-world surveillance.

But requiring that company, merely to complete its application for a CPT code, to characterize the possible “perpetuation” or “propagation” of “social bias” is something different.

AMA can obtain every medically important fact it needs without requiring that excursion.

At a time when the Secretary of HHS is already publicly challenging AMA's stewardship of CPT, continuing to place a mandatory “Social Bias” inquiry in the coding application looks less like sophisticated future-proofing than like a remarkably avoidable collision between two political eras.

It is, in short, an unforced error at almost the worst possible moment.