terminal / blog

Healthcare’s federal AI nutrition label is dead. Who reads it now?

November 2, 2026 · ai

What changed this week with HTI-5

Last week the Office of the National Coordinator for Health IT finalized HTI-5, removing the requirement for AI “nutrition labels” in certified EHR’s. Starting 2027, EHR vendors are no longer required to show users the source attributes behind their predictive models. The change - though it may sound scary - didn’t eliminate the obligation to examine the predictive models provided. It simply shifted the responsibility from the vendor to the customer.

HTI-1, created the “decision support interventions” certification criterion. Any predictive model provided by a certified EHR provider had to expose 31 source attributes across different categories:

Intended use and population Out-of-scope uses Training data Fairness approach External validation Performance Ongoing maintenance

This only applied to AI decision support built by the EHR developer itself. Third-party AI was never in scope and therefore most of the AI an organization purchases was not subject to HTI-1. There was never a federal guarantee of transparency for those tools in the first place.

Why ONC pulled it

ONC’s argument was that it found no public evidence the requirements improved patient care. The model transparency hadn’t delivered the benefits it expected. This decision was met with push back from clinical groups like the American College of Radiology who asked ONC to find another way to promote transparency in the absence of the requirement.

Another supporting argument for the requirements removal was the fact that no one was referencing these model cards buried 4 or 5 clicks deep in an EHR. It was a disclosure that mostly went unreferenced due to its poor positioning.

Who's Responsible now

In 2025, the Joint Commission and the Coalition for Health AI (CHAI) published guidance on responsible AI use. It advises healthcare organizations to require vendors to disclose the risks and limitations of each AI tool before it is deployed. Even though the certifications provided by CHAI and the Joint Commission are optional, they evaluate whether an organization is applying the principles of AI governance, bias assessment and vendor oversight (these voluntary accreditation programs have a way of becoming expectations).

The net effect of all this is that EHR vendors are no longer required to hand you a label with certification criteria. A healthcare organization must ask for it.