ONC’s HTI-5 Rule Drops AI Disclosure Rules for Coders

·

ONC’s HTI-5 Rule Drops AI Disclosure Rules for Coders

On October 26, 2026, the Office of the National Coordinator for Health IT (ONC) published the final HTI-5 rule, cutting 14 of the 60 certification criteria in the ONC Health IT Certification Program and revising seven more. ONC estimates the changes will save the industry $1.5 billion in compliance costs. Buried inside that deregulatory package is a change that matters directly to coding and compliance teams: the rule eliminates the “model card” transparency requirements that applied to AI-enabled decision support interventions, the category that covers most predictive and AI-assisted coding, CDI, and chart-review tools embedded in certified EHRs.

For a profession that has spent the last two years being told to validate AI output before trusting it, removing the federal requirement that vendors disclose how that AI was trained and tested is a significant shift, even if it arrived with little fanfare.

What HTI-5 Actually Removes

HTI-5 narrows the decision support intervention (DSI) certification criterion that ONC created under the 2023 HTI-1 rule. That criterion required developers of predictive and AI-enabled DSIs, including coding-suggestion engines, CDI prompts, and automated chart-review tools, to publish model-card-style documentation: source attributes, intended use, and the training, performance, and validation data behind the tool. Developers also had to support annual review of those disclosures.

The Criterion Coders Actually Relied On

That source-attribute and access-and-modification documentation was the mechanism coders, CDI specialists, and compliance officers used to answer a basic question before trusting an AI suggestion: what was this model trained on, and how was it validated? HTI-5 removes that requirement entirely rather than revising it, according to Healthcare Dive’s coverage of the final rule. ONC’s rationale, in the words of one official quoted in that coverage, is to “significantly lower the administrative compliance burden for developers.”

What Stayed in Place

ONC did not finalize its separate proposal to cut privacy and security criteria, including the capability that helps providers meet HIPAA obligations. Proposed information-blocking changes that would have extended the definitions of “access” and “use” to cover autonomous AI systems also remain under consideration rather than finalized. So the rollback is targeted: it hits AI transparency specifically, not certification broadly.

Why This Lands Differently for Coding Teams

Evergreen AI governance advice tells compliance teams to ask vendors for validation data before adopting a coding or CDI tool. HTI-5 doesn’t ban that disclosure, but it removes the federal floor that made it a certification requirement rather than a negotiated favor. In practice, that changes several things coding and compliance teams should expect going into 2027:

  • Vendors of certified EHR-embedded coding and CDI tools are no longer required to publish training, performance, or validation data as a condition of certification.
  • The annual review obligation tied to those disclosures goes away, so there’s no recurring federal checkpoint forcing an update when a model is retrained.
  • Coding compliance programs that cited ONC certification as evidence of AI due diligence will need a different basis for that claim in an audit.
  • Procurement and vendor-risk teams inherit the work of requesting model documentation directly, since it’s no longer a baseline certification condition.

The Burden Shifts to Health Systems and Vendors

Hospital and clinician commenters pushed back on the proposal for exactly this reason: removing a standardized disclosure requirement doesn’t remove the need to vet AI, it just moves that work from a federal checkpoint to each individual provider organization. For a health system running several AI coding or CDI tools across different vendors, that means building internal processes to request and evaluate the same information ONC used to require as a matter of course, with no common format to compare vendors against.

ONC has signaled it may revisit AI oversight in a future HTI-6 rulemaking, but has not committed to restoring transparency requirements or said when that rule might appear. For now, coding compliance programs are operating in the gap between HTI-1’s old standard and whatever, if anything, replaces it.

What Coding and Compliance Teams Should Do Now

Teams that have been leaning on ONC certification as a proxy for AI vetting should treat that proxy as gone for any coding or CDI tool certified or recertified after HTI-5 takes effect. The practical fix isn’t complicated, but it does require deliberate effort: build model documentation requests into vendor contracts directly, rather than assuming certification covers it. Ask specifically for training data provenance, validation methodology, and a commitment to notify customers when the underlying model changes. None of that is new advice, but it’s now the primary line of defense rather than a backstop behind a federal requirement.

Coding leaders should also expect this to come up in audits. If a compliance program’s documentation still cites “ONC-certified AI disclosure” as evidence of vendor vetting, that language needs updating to reflect what HTI-5 actually requires going forward, not what HTI-1 used to require.

Automated review has a role to play here too: a platform that logs and explains its own coding decisions internally gives a compliance team something to point to even when a vendor’s external disclosures get thinner. That’s part of what Medikode’s automated medical coding platform is built to provide, transparent, auditable coding decisions that don’t depend on a certification rule to be reviewable.