HHS drops AI model card rule for EHRs, cites $1.55 billion

HHS projects $1.55 billion in savings for health IT developers from its HTI-5 final rule, published in the Federal Register on Oct. 28. One change ends the rule that certified electronic health record software must let clinicians see how its AI and other decision tools were built and tested.

Those disclosures, often called model cards, covered a tool’s intended use, training data, limitations, performance and known risks. They came in with the HTI-1 rule of January 2024. The Office of the National Coordinator for Health IT announced the final rule on Oct. 26. It takes effect Dec. 28, with some pieces on Jan. 1, 2027.

Why it matters: hospitals that use AI built into their EHR will no longer get those details as a condition of the vendor’s federal certification. They will have to ask.

What the rule cuts

Of the 60 certification criteria in force, ONC proposed last year to remove 34 and revise seven. The final rule removes 14, revises seven and leaves 39 alone, per the ONC list.

The AI change sits in the “decision support interventions” criterion. The rule deletes three parts of it: source attribute support, source attribute access and modification, and intervention risk management. Other removals include the old clinical decision support criterion, care plan, family health history and both Direct Project messaging criteria.

The savings figure is in 2024 dollars at a 7% discount rate, or $108.1 million a year. HHS estimates up to 86% of it, about $93 million a year, could go to small developers. Its count is 466 developers of certified health IT.

Why ONC says the disclosures had to go

In the rule, ONC says developers told it “most clinical users lack the time or technical background” to use the source attribute details. It adds that it found no evidence in peer reviewed research or the trade press that the disclosures “led to positive impacts on patient care.”

“American doctors and patients need better technology, not more Washington paperwork,” HHS Secretary Robert F. Kennedy Jr. said in the release.

ONC also says it “will take under consideration for future rulemaking” whether some of the transparency rules should come back.

What commenters warned

The rule summarizes the pushback it got. Commenters said the cut would shift the work and liability of vetting AI tools to hospitals and doctors. They said small, rural, safety net and long term care providers would feel it most, since many lack data science and legal staff. Others said it would make bias and performance harder to check and could raise the risk of clinicians leaning too hard on automated advice.

ONC’s answer: users should ask their vendor for that information before they turn a tool on. It also reminds providers they still must identify bias risk in such tools under the federal nondiscrimination rule at 45 CFR 92.210. A feedback requirement that lets providers export and report how evidence based tools perform in real use stays in place.

What to watch

The rule leaves the HTI-5 information blocking proposals out. ONC says comments on those “remain under consideration.” For hospitals, vendor contracts and AI governance reviews now become the main place to get the facts the certification rule used to require.