A computer workstation on a wheeled grey cart, its monitor dark and an IV pole clamped to its side, stands against the wall of a dim hospital corridor at night, the polished floor reflecting a few ceiling lights and closed double doors at the far end.

Health AI

HHS Will Stop Requiring ‘Nutrition Labels’ for Health-Record AI

HHS ends the 2024 requirement that certified health-record software explain its predictive AI tools; it puts developers’ savings at $7.4 million to $53.3 million a year in labor.

WASHINGTON, Oct. 28 — The Department of Health and Human Services will no longer require certified health-record software to show its users how the predictive artificial-intelligence tools built into it were trained, tested and checked for bias. A final rule published in Wednesday’s Federal Register strikes the 2024 requirements for these “model cards,” which it likens to nutrition labels, along with a duty on developers to manage the tools’ risks. The decision-support changes take effect Dec. 28.

The rule comes from the department’s Office of the National Coordinator for Health Information Technology, or ONC, which runs a voluntary program that certifies health-information technology. Of the program’s 60 certification criteria in effect, it removes 14 and revises seven; its proposal, on Dec. 29, 2025, would have removed 34. One of the “more substantial” revisions, it says, is “to fully remove the super intelligence ‘model card’ requirements.” The rule calls artificial intelligence “super intelligence,” or SI, citing an executive order of Sept. 29.

The 2024 rule, known as HTI-1, required software certified for “decision support interventions” to let users see “source attributes” for each predictive tool its developer supplied: among them its intended use and patients, its known risks and limitations, the criteria that shaped its training data, how it was validated on outside data, measures of its validity and fairness, and how it is monitored. For some of the validation and monitoring fields, developers could instead indicate that the information was not available. Developers also had to analyze and mitigate each tool’s risks, including risks to its validity, fairness, safety and privacy.HTI-1The 2024 rule was published Jan. 9, 2024, over the signature of Xavier Becerra, then the health secretary. From Jan. 1, 2025, its decision-support criterion replaced the older clinical decision support criterion in the program’s definition of a base electronic health record.

By ONC’s count, 251 products from 206 developers were certified to the decision-support criterion. ONC estimates that the broader revision of the decision-support criterion, which also drops disclosures for evidence-based tools, will save developers $7.4 million to $53.3 million a year in compliance labor, before discounting. Developers get the savings; commenters warned that the work of vetting the tools, and the liability, would pass to hospitals and health systems.

“A large majority of commenters were not supportive of our proposal,” the rule says, “primarily due to concerns about removing DSI transparency and risk-management requirements.” Many called the move premature, with the requirements in effect for less than two years; removing them, commenters said, “would create a significant regulatory gap in SI governance at a time when SI adoption is accelerating.” Some said the shift would be especially hard on small, rural and safety-net providers. Supporters noted “extremely low use” of the functions, and several said industry-led frameworks were better suited than federal rules that could quickly become outdated.

Rob Bonta, California’s attorney general, wrote in a comment letter on the proposal on Feb. 27 that dropping the requirement “would eliminate one of the most significant guardrails currently in place on a federal level for the use of AI in healthcare.” Many hospitals his office had queried, he wrote, “relied on the vendor’s assessment that the tools they use are ethical and unbiased.”

ONC was unmoved. “We believe that we have had a sufficient amount of time to evaluate the effectiveness of the transparency requirements,” it wrote, and said it was “not persuaded of their value, particularly given the feedback we have received from developers that the requirements are burdensome and detrimental to innovation.” The agency said it had found no evidence in peer-reviewed research or the trade press that users relied on the labels or that they improved patient care, and judged any significant shift of burden to providers “unlikely.”

Certified software must still let users switch on both evidence-based and predictive tools, and collect feedback on the evidence-based ones. ONC reminded users of a separate rule, 45 CFR 92.210, against discrimination through decision-support tools.Section 92.210Under the department’s rule for the Affordable Care Act’s anti-discrimination section, a covered entity, such as a recipient of federal financial assistance, must not discriminate on the basis of race, color, national origin, sex, age or disability through patient care decision support tools, and must make “reasonable efforts” to identify tools that use such factors as inputs and to mitigate the risk of discrimination from them. It said it would consider, in future rulemaking, whether to reinstate older source-attribute requirements for evidence-based tools; that passage does not cover the model cards for predictive ones.

Some commenters warned that removing the federal framework could produce “divergent State requirements.” For now, ONC’s advice to hospitals is to ask. “We encourage users of certified health IT to ask their developers for such information,” the rule says.

Sources

  1. U.S. Department of Health and Human Services, Office of the National Coordinator for Health Information Technology (Federal Register, 91 FR 67916), “Health Data, Technology, and Interoperability: ONC Deregulatory Actions To Unleash Prosperity (final rule),” Oct. 28, 2026.
  2. Federal Register, “Full text: Health Data, Technology, and Interoperability: ONC Deregulatory Actions To Unleash Prosperity (FR Doc. 2026-21943),” Oct. 28, 2026.
  3. U.S. Department of Health and Human Services (Federal Register, 89 FR 1192), “Health Data, Technology, and Interoperability: Certification Program Updates, Algorithm Transparency, and Information Sharing (HTI-1 final rule),” Jan. 9, 2024.
  4. Electronic Code of Federal Regulations, “45 CFR 170.315: ONC certification criteria for health IT (paragraph (b)(11), decision support interventions, as in force before Dec. 28),” Oct. 26, 2026.
  5. Electronic Code of Federal Regulations, “45 CFR 92.210: Nondiscrimination in the use of patient care decision support tools,” Oct. 26, 2026.
  6. California Department of Justice, Office of the Attorney General, “Comment letter of Attorney General Rob Bonta on the proposed rule (RIN 0955-AA09),” Feb. 27, 2026.
  7. Federal Register, “Health Data, Technology, and Interoperability: ASTP/ONC Deregulatory Actions To Unleash Prosperity (proposed rule, 90 FR 60970),” Dec. 29, 2025.
  8. Federal Register, “Executive Order 14434: Inaugurating the Era of Super Intelligence,” Oct. 2, 2026.