HL7 AI Transparency on Fast Healthcare Interoperability Resources (FHIR)

Overview

Health data is increasingly produced or modified by AI, but the record itself rarely contains this information. To downstream users, AI-influenced data included in conventionally produced health records, are indistinguishable from other data, and AI involvement cannot be determined from the data itself.

AI Transparency on Fast Healthcare Interoperability Resources (FHIR) is a Health Level Seven (HL7) International implementation guide that aims to address this issue by standardizing how AI involvement is recorded in health data exchanged using FHIR. It defines two complementary levels of transparency:

• Lightweight coded tag — marks a resource, or a single element within a resource, as AI-influenced
• Fuller Provenance record — identifies the AI system, the human and automated participants and the roles they played, the inputs and prompts, and a link to the system's model card

Notably, the guide standardizes representation only; it does not evaluate AI systems, judge the resulting data, or detect undisclosed AI use.

NIST contributes to authoring and technical review of the implementation guide. Researchers in the Applied AI Group (Division 775.02) of the Information Technology Laboratory (ITL) AI Program provide measurement science expertise to help ensure the standard enables meaningful transparency without imposing unworkable burdens on health IT systems.

Objectives

• Standardize how AI involvement is recorded in health data exchanged via FHIR
• Define lightweight and detailed mechanisms for flagging AI-influenced resources
• Enable downstream users of health data to distinguish AI-produced or AI-modified content from human-authored content
• Support interoperability across health IT systems by building on FHIR version R4 with forward compatibility to versions R5 and R6

Approach

The implementation guide is developed through HL7 International's consensus-based standards process. The guide is built on FHIR R4 and is forward compatible with R5 and R6. The first draft of the guide has been publicly reviewed through HL7’s formal balloting process and approved for trial use as an STU1 ballot draft. Healthcare organizations may therefore begin piloting and implementing the standard. However, additional revisions and rounds of review of the guide are expected before it is finalized.

The standard defines two complementary transparency mechanisms. The first is a simple coded tag that can be applied at the resource or element level, providing a minimal-overhead signal that AI was involved. The second is a structured Provenance resource that captures richer detail: the identity of the AI system, the roles of human and automated participants, the inputs and prompts used, and a reference to the system's model card.

Resources

The implementation guide is developed through HL7 International, the global authority for healthcare data interoperability standards. The guide is written for the FHIR (Fast Healthcare Interoperability Resources) standard, which defines how health information is structured and exchanged between electronic health systems.

STU1 Ballot Draft — The version that completed the first round of public review through HL7's formal balloting process. This is the most recent community-reviewed edition of the guide.

Current Build — The working draft that reflects ongoing revisions since the last ballot. This version includes the latest edits but has not yet been formally reviewed.