AI Regulation Tracker / Proposed rule

The AI Model Card in Your EHR Could Vanish: ONC Proposes Repealing the HTI-1 DSI Transparency Rule

A new federal proposal would strip out the certified-EHR requirement to show clinicians how a predictive AI tool was built and validated. It is a proposed rollback, not law. The HTI-1 model-card requirements stay binding unless and until a final HTI-5 rule changes them, and the analysis below assumes the proposal is adopted as written.

By Anthony Guerriero · Reviewed by The Leveraged Years Editorial Desk · Published June 27, 2026 · Last updated June 27, 2026

The AI Model Card in Your EHR Could Vanish: ONC Proposes Repealing the HTI-1 DSI Transparency Rule regulation briefing

The Leveraged Years AI Regulation Tracker

Regulatory Briefing

Bindingness: Non-Binding Framework (proposed rollback of an in-force rule)
Scope: United States (federal) / Sector: certified health IT, predictive AI in electronic health records
What Changed: The Assistant Secretary for Technology Policy and Office of the National Coordinator (ASTP/ONC) issued the HTI-5 proposed rule, which would reduce the scope of the Decision Support Intervention (DSI) certification criterion and remove the predictive DSI "source attribute" and intervention risk management requirements, the provisions clinicians know as the AI "model card."
Who This Affects: Physicians and clinicians who consult AI-generated risk scores, flags, or recommendations inside certified EHRs; the informaticists and health-IT leaders who configure those systems; and the developers who today must publish source-attribute information to keep a certified DSI module.
Current Status: Proposed (NPRM). Not in effect.
Effective Date: None. Comment period closed February 27, 2026. The current HTI-1 DSI transparency requirements remain in force unless and until a final rule changes them.
Primary Source: Federal Register, "Health Data, Technology, and Interoperability: ASTP/ONC Deregulatory Actions To Unleash Prosperity" (HTI-5 proposed rule), published December 29, 2025, document 2025-23896: https://www.federalregister.gov/documents/2025/12/29/2025-23896/health-data-technology-and-interoperability-astponc-deregulatory-actions-to-unleash-prosperity
Secondary Analysis: Holland and Knight, health-IT alert on the HTI-5 proposed rule and DSI changes.
Workflow Impact: Today, under the in-force HTI-1 rule, a certified EHR has to surface source-attribute details for a predictive intervention, the intended use, the patient population, known risks, the training data described in general terms, and how the tool was validated. Those requirements remain binding now. If HTI-5 is finalized as proposed, a developer would no longer need to provide that panel to stay certified, so the transparency layer a clinician was trained to check before trusting an AI output could become optional from a certification standpoint. This analysis assumes HTI-5 is adopted as proposed; the agency could change or drop the provision before any final rule.
Operational Next Step: Do not build a clinical habit on the assumption that the model-card data will always be one click away. Confirm with your informatics team which AI tools currently expose source attributes, capture that information now, and decide how your group will vet a predictive tool if the certified disclosure goes away.
Last Verified: June 27, 2026

The transparency panel you were just taught to read may not survive

Over the past two years, clinicians have been told to do something new before trusting an AI suggestion in the chart. Look at the model card. Check what the tool was built for, which patients it was trained on, and how it was validated. That panel exists because a federal rule put it there. A new federal proposal would take it back out.

On December 29, 2025, the Assistant Secretary for Technology Policy and Office of the National Coordinator, ASTP/ONC, published a proposed rule in the Federal Register titled "Health Data, Technology, and Interoperability: ASTP/ONC Deregulatory Actions To Unleash Prosperity," known as HTI-5 (document 2025-23896). The comment period closed on February 27, 2026. Among dozens of changes to the certification program, one matters directly to anyone who reads an AI output at the bedside: the proposal would reduce the scope of the Decision Support Intervention criterion to, in the agency's own words, "fully remove the artificial intelligence (AI) model card requirements."

What is in force today, and what is only proposed

First, the important caveat. This is a proposed rule, not law. The transparency requirements clinicians rely on come from an earlier rule, HTI-1, which is in force now. HTI-1 created the Decision Support Intervention certification criterion and required certified health IT to surface "source attributes" for predictive interventions, the information the field started calling the AI model card. Those requirements remain binding today.

HTI-5 is the sequel that would unwind part of that. As proposed, it would remove the predictive DSI source attribute and intervention risk management provisions from the certification criterion, which the agency describes as a move to "fully remove the artificial intelligence (AI) model card requirements." Nothing has changed yet. A proposed rule is the government showing its intent and asking for comment, and the agency can finalize it, change it, or drop it. The HTI-1 model card requirements remain binding and in effect until and unless a final HTI-5 rule is published, and the analysis here assumes HTI-5 is adopted as proposed. Until a final rule appears, the model card stays where it is. The reason to pay attention now is that the direction of travel is set, and it points away from mandatory disclosure.

What the source attribute actually gave clinicians

It helps to be concrete about what could go away. Under the current HTI-1 criterion, a certified EHR has to make available, for a predictive intervention, a defined set of source-attribute fields. These describe the intended use of the tool, the patient population and intended users, known risks and limitations, a plain description of the data and inputs the tool relies on, and the process used to validate it, including external validation where it applies. Developers also have to attest to intervention risk management practices.

For a physician, that set of fields is the difference between a number that appears in the chart and a number you can interrogate. It is how you learn that a sepsis or readmission score was validated on a population that may or may not resemble the patient in front of you. HTI-5 would make providing that information optional from a certification standpoint, which means a vendor could keep it, simplify it, or remove it.

Why a deregulatory proposal is still a clinical-safety issue

The agency frames HTI-5 as cutting requirements that are duplicative or that slow innovation, and a vendor that values the model card may keep publishing it voluntarily. That is the optimistic read. The clinical risk is quieter. If the disclosure is no longer required, it will not disappear all at once or with an announcement. It may thin out, move, or vanish from one product and not another, so two clinicians using two certified systems could see very different levels of transparency for the same kind of AI tool.

This is the part worth saying plainly to a clinical audience. A workflow habit is only as durable as the rule underneath it. If your protocol for trusting an AI flag depends on a panel that a future final rule could make optional, the protocol needs a backup that does not assume the panel will be there.

What to do while it is still proposed

The practical work is documentation and habit. Ask your informatics or EHR governance team to list the predictive tools running in your system and to capture, today, the source-attribute details each one currently exposes. Save that information outside the vendor's optional panel so your group keeps it even if the disclosure is later removed. Decide now how a physician should vet a new AI tool if the certified model card is gone, what questions to ask the vendor, and what evidence to require before relying on an output. And if your organization commented before the February 27, 2026 deadline, track where the rule goes from here, because the move from proposed to final is the moment the chart actually changes.

Source File

Primary Source: Federal Register, "Health Data, Technology, and Interoperability: ASTP/ONC Deregulatory Actions To Unleash Prosperity" (HTI-5 proposed rule), December 29, 2025, document 2025-23896: https://www.federalregister.gov/documents/2025/12/29/2025-23896/health-data-technology-and-interoperability-astponc-deregulatory-actions-to-unleash-prosperity
Secondary Source: Holland and Knight, health-IT alert on the HTI-5 proposed rule and the Decision Support Intervention changes.
Last Checked: June 27, 2026
Status: Proposed (NPRM). Comment period closed February 27, 2026. HTI-1 DSI transparency requirements remain in force.
Editorial Note: This is workflow-level analysis for physicians and health-IT teams, not medical or legal advice. Confirm current certification obligations and the status of any final rule with qualified counsel and your informatics leadership.

Frequently Asked Questions

Has the AI model card requirement been removed from certified EHRs?

No. As of June 27, 2026, the model card requirement is still in force under the HTI-1 rule. The HTI-5 proposed rule, published December 29, 2025, would remove it, but it is a proposed rule and not law. The current requirement stays in place unless and until a final rule changes it.

What is the DSI source attribute or AI model card?

It is the set of disclosure fields a certified EHR must surface for a predictive Decision Support Intervention, including the intended use, the patient population and users, known risks, a description of the data, and how the tool was validated. HTI-5 proposes making these optional for certification.

When did the HTI-5 comment period close?

The comment period for the HTI-5 proposed rule closed on February 27, 2026.

Should clinicians stop relying on the model card?

Not while it is still present and required. The point for physicians is to avoid assuming it will always be there. Capture the source-attribute information now and build a way to vet AI tools that does not depend on a disclosure a future final rule could make optional.