Posted on
Apr 4, 2026
ONC Health IT Certification for AI Scribes: What EHR Administrators Must Know in 2026
ONC Health IT Certification for AI Scribes: What EHR Administrators Must Know in 2026
As AI-powered clinical documentation tools become embedded in daily workflows across U.S. health systems, EHR administrators face a pressing question: does the AI scribe integrated into your certified health IT stack meet federal certification requirements? Platforms like Scribing.io are building their AI documentation tools with these evolving regulatory frameworks in mind—but the compliance burden ultimately falls on the organizations deploying them.
The landscape shifted significantly when the Assistant Secretary for Technology Policy (ASTP)/ONC finalized the HTI-1 Final Rule, establishing for the first time that predictive AI algorithms—including ambient AI scribes—must meet specific transparency, risk management, and fairness criteria when operating within certified EHR systems. For EHR administrators responsible for procurement, compliance, and federal mandate adherence, understanding these requirements is no longer optional. This guide provides a detailed breakdown of what the certification criteria mean, what you need to verify from vendors, the compliance timeline, and how Scribing.io approaches these standards.
Important: This page provides regulatory education for health IT professionals. It does not constitute legal advice. The ASTP/ONC Health IT Certification Program now includes specific requirements for AI-powered clinical tools—including ambient AI scribes—under the Predictive Decision Support Intervention (DSI) criterion established by the HTI-1 Final Rule. EHR administrators must verify that any AI scribe integrated into their certified health IT stack meets transparency, risk management, and fairness requirements or risk non-compliance with federal mandates. This guide breaks down what the certification criteria mean, what EHR administrators need to verify, the compliance timeline, and how Scribing.io approaches these standards. Organizations should consult qualified health IT legal counsel for institution-specific compliance decisions.
Table of Contents
What Is ONC Health IT Certification and Why Does It Now Apply to AI Scribes?
The HTI-1 Final Rule Explained: Predictive DSI Criteria for AI Documentation Tools
What EHR Administrators Must Verify Before Deploying an AI Scribe
ASTP/ONC Certification vs. FDA Regulation: Understanding the Jurisdictional Boundaries
How Scribing.io Approaches ONC Compliance and Responsible AI
Get Started Today
What Is ONC Health IT Certification and Why Does It Now Apply to AI Scribes?
The ONC Health IT Certification Program—now administered by the Assistant Secretary for Technology Policy (ASTP) within the Department of Health and Human Services—serves as the federal government's primary mechanism for ensuring that electronic health record technology meets baseline standards for interoperability, security, and functionality. Certified EHR technology (CEHRT) is required for participation in CMS programs including the Medicare Promoting Interoperability Program and MIPS. The vast majority of U.S. hospitals and physician practices use ONC-certified systems, making the certification program a de facto regulatory gatekeeper for health IT adoption.
Historically, ONC certification focused on structured data standards, interoperability requirements, and clinical quality measure reporting. AI-powered tools operating within or alongside EHR platforms existed in a regulatory gray zone—acknowledged but not explicitly addressed by certification criteria. That changed with the publication of the HTI-1 Final Rule, which for the first time created a distinct certification pathway for predictive algorithms and AI-powered tools, including ambient AI scribes that generate clinical documentation from patient-clinician conversations.
Why the Distinction Between EHR Certification and AI Module Certification Matters
A critical nuance for EHR administrators: the ONC certification of your EHR platform (e.g., Epic, athenahealth, Cerner) does not automatically extend to every third-party AI module integrated into that platform. An AI scribe that operates as a Predictive Decision Support Intervention within a certified EHR must independently satisfy the Predictive DSI certification criteria—or the EHR developer must include that AI functionality within the scope of its own certification and accept responsibility for meeting those criteria.
This distinction has significant procurement implications. When ModMed announced in January 2025 that it had achieved ONC certification specifically for its AI scribe capabilities, it signaled that the market was moving toward treating AI scribe certification as both a competitive differentiator and a compliance expectation. EHR administrators should expect vendors to increasingly be asked to demonstrate certification status or a credible roadmap toward it. Explore how Scribing.io's AI scribe features are designed with compliance in mind.
The HTI-1 Final Rule Explained: Predictive DSI Criteria for AI Documentation Tools
The HTI-1 Final Rule, effective March 11, 2024, represents the most significant expansion of ONC's certification framework in over a decade. While the rule addresses multiple areas of health IT regulation, four pillars are most relevant to organizations deploying AI tools:
Algorithm Transparency: New requirements for disclosing how AI and predictive models function within certified health IT.
USCDI v3 Adoption: Updated data standards (United States Core Data for Interoperability, Version 3) that certified systems must support, with a compliance date of January 1, 2026.
Enhanced Information Blocking Requirements: Strengthened provisions to prevent interference with the exchange of electronic health information (EHI).
The Insights Condition: New reporting obligations for certified health IT developers regarding how their products are used in real-world settings.
Decision Support Interventions: Predictive vs. Evidence-Based
The section of HTI-1 most directly relevant to AI scribes is the Decision Support Interventions (DSI) certification criterion, codified at § 170.315(b)(11). ONC draws a critical distinction between two categories:
Predictive DSI encompasses any intervention that uses algorithms, machine learning, or other computational methods to generate outputs that are not purely derived from established clinical guidelines or peer-reviewed literature. Ambient AI scribes—which use large language models to interpret spoken clinical encounters and generate draft documentation—fall squarely into this category because their outputs are model-generated rather than rule-based.
Evidence-based DSI includes traditional clinical decision support such as drug-drug interaction alerts, evidence-based order sets, and guideline-driven reminders. These interventions derive their outputs from codified clinical evidence.
What the Predictive DSI Criterion Requires
For any AI scribe seeking to operate within the ONC certification framework, the Predictive DSI criterion imposes three major categories of requirements:
Source Attribute Disclosures: Developers must publish detailed information about their AI models, including descriptions of training data, the model type and architecture, performance metrics (such as accuracy, precision, and recall where applicable), known limitations, and the intended use population. These disclosures must be accessible to both the clinician end-user and the organization deploying the tool.
Intervention Risk Management (IRM): Developers must maintain a documented governance framework that includes risk identification and mitigation practices, ongoing model monitoring, defined roles and responsibilities, staff training, and procedures for addressing model drift or degraded performance over time.
FAVES Framework: ONC introduced the FAVES principles—Fairness, Appropriateness, Validity, Effectiveness, and Safety—as the evaluative lens through which Predictive DSI should be assessed. While FAVES is currently a framework rather than a strict pass/fail test, it establishes the expectations that ONC-Authorized Certification Bodies (ONC-ACBs) will use when evaluating AI tools for certification.
See how these federal rules intersect with state-level AI scribe regulations like California's.
Predictive DSI vs. Evidence-Based DSI: Requirements Comparison
Requirement | Predictive DSI (AI Scribes) | Evidence-Based DSI |
|---|---|---|
Source attribute disclosure | Required: training data, model type, performance metrics, limitations, intended population | Required: bibliographic citation, developer, funding source, clinical evidence basis |
Intervention Risk Management program | Required: governance framework, monitoring, bias evaluation, staff training | Not required under DSI criterion |
FAVES evaluation | Required: Fairness, Appropriateness, Validity, Effectiveness, Safety | Not explicitly required |
User-facing disclosure at point of care | Required: clinician must be able to access source attributes | Required: bibliographic references must be accessible |
Ongoing monitoring obligation | Required: continuous performance and bias monitoring | Periodic review recommended but not mandated |
USCDI v3 structured data output | Required (effective January 1, 2026) | Required (effective January 1, 2026) |
What EHR Administrators Must Verify Before Deploying an AI Scribe
For EHR administrators evaluating AI scribe vendors, the HTI-1 framework creates a concrete set of verification steps. The following checklist translates the regulatory requirements into actionable procurement and compliance criteria:
Certification status: Is the AI scribe module listed on the ONC Certified Health IT Products List (CHPL)? If not, does the vendor have a documented certification roadmap with specific milestone dates? Verify whether the AI scribe is certified independently or covered under the EHR developer's broader certification.
Predictive DSI transparency disclosures: Has the vendor published the required source attributes? Request documentation of data provenance, model architecture descriptions, performance benchmarks, known limitations, and the intended clinical use case. Incomplete or absent disclosures are a red flag.
Intervention Risk Management (IRM) program: Does the vendor maintain a documented governance framework? Ask for evidence of defined policies and procedures, named responsible individuals, staff training records, third-party audit or monitoring capabilities, and incident response protocols for model failures.
USCDI v3 compliance: Does the AI scribe output structured data aligned with USCDI v3 standards? This is particularly important for organizations that need AI-generated notes to populate discrete EHR fields rather than exist solely as free-text blobs.
Information blocking provisions: Does the AI scribe's architecture support the exchange of electronic health information under TEFCA, or does it create proprietary data silos that could constitute information blocking?
Insights Condition reporting: Can the vendor support the certified health IT developer's obligation to report real-world usage data under the Insights Condition? This is especially relevant for AI scribes integrated into certified EHR platforms where the EHR developer bears reporting responsibility.
Bias and equity evaluation documentation: Has the vendor conducted and disclosed fairness assessments across patient demographics including race, ethnicity, age, sex, and language? Ask for specific methodology descriptions, not just general statements of commitment to equity.
Ongoing update and re-certification plan: What is the vendor's roadmap for maintaining certification as ASTP/ONC rules evolve? The anticipated HTI-2 rulemaking is expected to further refine Predictive DSI requirements, and vendors without a clear regulatory strategy may leave your organization exposed.
ASTP/ONC Certification vs. FDA Regulation: Understanding the Jurisdictional Boundaries
One of the most common sources of confusion among EHR administrators is the relationship between ONC certification and FDA regulation of AI tools. These are separate regulatory frameworks with different scopes, different enforcement mechanisms, and different implications for your organization.
ONC/ASTP Certification
ASTP/ONC certification governs health IT functionality, interoperability, and transparency for tools operating within certified EHR systems. It is primarily a technology standards framework. Certification ensures that AI tools meet disclosure, governance, and data standards requirements. Non-compliance with ONC certification requirements can affect an organization's eligibility for CMS incentive programs and expose the health IT developer to information blocking penalties.
FDA Oversight
The FDA regulates clinical decision support software that meets the definition of a medical device under Section 3060 of the 21st Century Cures Act. Under the FDA's CDS guidance, software qualifies for FDA oversight when it is intended for use in the diagnosis, treatment, or prevention of disease and either acts autonomously (without clinician review) or provides patient-specific recommendations that clinicians are unlikely to independently verify.
Where AI Scribes Currently Fall
Most ambient AI scribes generating draft clinical documentation are designed as clinician-review tools—the AI produces a draft note that the clinician reviews, edits, and signs. This workflow positions them under ONC's Predictive DSI framework rather than as FDA-regulated devices, because the clinician retains decision-making authority over the final documentation.
However, this boundary is fact-specific and evolving. An AI scribe that begins making autonomous clinical recommendations—such as suggesting diagnoses, recommending medication changes, or triggering clinical orders based on conversation analysis—could cross into FDA jurisdiction. The regulatory line depends on the tool's intended use and the degree of clinician intermediation in the workflow.
A growing body of evidence shows AI scribes are becoming deeply embedded in mainstream clinical workflows. Research published in JAMA has documented the expansion of AI documentation tools across major academic medical centers, making regulatory clarity even more critical for administrators who must ensure their deployed tools are governed by the correct framework.
Learn how AI scribes integrate with Epic and what that means for certification.
For EHR administrators, the practical takeaway is this: verify which regulatory framework applies to each AI tool in your stack. An AI scribe that is appropriately governed under ONC certification today could trigger FDA jurisdiction if its feature set expands. Procurement contracts should address this contingency explicitly, and legal counsel should review edge cases where the tool's functionality approaches the FDA's CDS criteria.
How Scribing.io Approaches ONC Compliance and Responsible AI
At Scribing.io, the HTI-1 framework is not treated as a checkbox exercise but as a foundational design principle. The platform's approach to ONC compliance spans model documentation, governance infrastructure, data standards, and ongoing monitoring.
Transparency and Source Attribute Documentation
Scribing.io maintains detailed model documentation aligned with the Predictive DSI source attribute requirements. This includes descriptions of training data composition, model architecture and version history, performance benchmarks relevant to clinical documentation accuracy, known limitations and edge cases, and intended clinical use populations. This documentation is designed to be accessible to both deploying organizations and clinician end-users, fulfilling HTI-1's user-facing disclosure mandate.
Internal Governance and Quality Assurance
Scribing.io operates an Intervention Risk Management program that includes defined governance roles and responsibilities, documented policies and procedures for model development, testing, and deployment, ongoing quality assurance review of AI-generated clinical notes, and incident response protocols for identified model errors or performance degradation. Clinicians who use the platform report that the human-in-the-loop review workflow—where every AI-generated note requires clinician sign-off—serves as both a quality safeguard and a regulatory compliance mechanism.
USCDI v3 Structured Data Output
Scribing.io's AI scribe is designed to output documentation in structured formats compatible with USCDI v3 standards. This is particularly important for organizations using the platform alongside certified EHR systems like athenahealth, where structured data output must flow cleanly into discrete EHR fields to support clinical decision-making, quality reporting, and interoperability requirements.
Ongoing Monitoring and Bias Evaluation
Scribing.io conducts ongoing performance monitoring that includes evaluation across patient demographic categories consistent with the FAVES framework's fairness requirements. The platform's monitoring infrastructure tracks documentation quality metrics, identifies potential performance variations across clinical specialties and patient populations, and feeds results back into model improvement cycles.
EHR Integration and Certification Posture
Scribing.io's integration architecture is designed to support—rather than complicate—an organization's overall ONC certification posture. By functioning as a documentation generation tool that passes structured data to the certified EHR, the platform minimizes the risk of creating information blocking scenarios or disrupting existing certification compliance. For specialty-specific workflows, the platform supports documentation patterns across cardiology, psychiatry, and other high-documentation specialties.
Transparency About Current Status
Scribing.io is actively engaged in the ONC certification process and is building its technology and governance infrastructure to meet Predictive DSI requirements. The platform's leadership has stated that achieving and maintaining ONC certification alignment is a strategic priority. Organizations evaluating Scribing.io should request the most current certification status documentation and roadmap milestones directly from the Scribing.io compliance team, as this landscape evolves rapidly.
For organizations navigating the intersection of AI documentation, ICD-10 coding, and federal compliance, Scribing.io's ICD-10 coding tools are designed with the same compliance-first philosophy applied to the broader platform.
Get Started Today
Federal certification requirements for AI scribes are no longer theoretical—they are active compliance obligations that EHR administrators must address in every procurement decision. Whether you are evaluating a new AI documentation tool or auditing your existing stack, the Predictive DSI criteria under HTI-1 provide a clear framework for what to demand from vendors. Scribing.io is built to meet these standards and to make compliance a seamless part of your AI-powered documentation workflow.
