Blog · Policy & safety

2025 ONC SAFER Guides put AI transparency on the EHR safety agenda

ONC’s Safety Assurance Factors for EHR Resilience (SAFER) Guides were updated to address AI-enabled systems. Here is what the public materials emphasize—and how research tooling can support transparency practice without claiming regulatory compliance.

What changed

ONC published updated 2025 SAFER Guides (building on the 2024 revision work) focused on high-priority EHR safety practices. A notable addition: recommended practices for AI-enabled applications—including transparency, inventory/oversight of AI tools, human review of AI outputs before they reach patients, and ongoing local performance monitoring (drift / bias / decay themes).

Canonical hub: healthit.gov — SAFER Guides · High Priority Practices PDF on HealthIT.gov.

ONC-certified health IT is widely used across U.S. hospitals and ambulatory settings; organizations that rely on certified EHR technology are the primary audience for SAFER self-assessment—not research software alone.

Themes that matter for AI risk modeling research

Transparency — know which AI tools are in use and for what purpose
Inventory & ownership — version, responsible owners, review dates
Human oversight — clinicians remain accountable for recommendations that affect care
Local evaluation — test on local data before routine use; monitor after deployment

How the EHR Risk Framework helps (methods alignment)

For labs and courses teaching trustworthy clinical ML methods:

Explainability practice — SHAP summaries and Predict explanations (when generated)
Performance honesty — Brier / ECE alongside discrimination metrics
Data integrity habits — leakage audits and index/horizon task specs
Reproducible packs — Docker + results ZIP for methods appendices

Sources

ONC. SAFER Guides. healthit.gov/topic/safety/safer-guides
ONC. SAFER High Priority Practices (PDF). Available via HealthIT.gov downloads.

Try the workbench

How it works A–Z Live demo Quickstart GitHub

Free demo server — it may be slow. Check it with a small amount of data. For larger workloads or freer experimentation, run locally or on your own server. Open live demo

Feedback: support@larucare.com · Cite: DOI & CITATION.cff