HealthRecordCommunity
FHIR🌏 InternationalEnriched

Five Caveats About an AI Agent Writing to a FHIR Record: Provenance and Regulatory Challenges

AI Agent EHR Write Access | FHIR - Pharos Production

August 27, 2026

Summary

When an AI agent writes information into electronic health records, several potential technical and legal issues exist at the specification level. Key concerns include proving authorship (Provenance) and managing permissions (Scope), highlighting that these are not just implementation details but also critical for regulatory compliance as a medical device.

Details

This article outlines technical and legal challenges associated with AI agent write processes into EHRs. Firstly, regarding Provenance—the proof of who wrote the data—while FHIR specifies an X-Provenance header, servers may ignore it. This means that merely attaching the header is only a claim about the agent's behavior, not a guarantee from the server. Secondly, SMART App Launch scopes are merely indications of allowed scope; they do not constitute absolute permission, meaning a write interaction could still return a 403 Forbidden error even if the scope appears granted. Thirdly, there can be discrepancies between the record creator (Agent) and the audit log recorder (Server), leading to ambiguity about accountability. Furthermore, because FHIR lacks signed status for clinical resources, it presents technical gaps in meeting transparency requirements like those mandated by the AI Act. These issues are not merely engineering choices; they directly relate to the documentation required in a technical file that must be presented to regulatory bodies (Notified Body) when classifying an AI system as a medical device, making them crucial considerations for healthcare IT adoption globally.

📰
Read Original Article
pharosproduction.com
→

Original content copyright by respective publishers