HTI-2 enforcement is coming: the Q1 2027 playbook for vendors and health systems

The ONC enforcement framework for HTI-2, published this month, converts a two-year-old rule into a near-term operational obligation. The framework is more rigorous than the industry’s working assumptions in three specific ways — it mandates quarterly attestation, it reaches past vendor documentation into live customer installations via random audits, and it requires public posting of the Decision Support Intervention (DSI) inventory for every certified installation. The combined effect is that compliance can no longer be managed as an annual documentation exercise. It is now a continuous operational discipline that reaches across the vendor, the health system, and the clinical workflow.

The core compliance target is the DSI source-attribute metadata set that HTI-2 defined back in 2024. Every predictive algorithm, every embedded decision support rule, every AI-enabled feature that influences a clinical decision inside a certified EHR has to have its source attributes documented and exposed. The source attributes include the model inputs, outputs, intended use, training-data characteristics, performance across subpopulations, uncertainty characteristics, and known limitations. In the pre-enforcement period, many vendors treated this requirement as a research-paper-style disclosure. The enforcement framework makes clear that the agency expects operational-grade metadata capable of informing a clinician’s live use of the tool.

What changed with the enforcement framework

The enforcement framework introduced three mechanisms that fundamentally change the compliance posture.

The quarterly attestation requirement means the certified vendor’s compliance officer is signing off every three months that the DSI inventory is current, that each listed DSI has complete source-attribute metadata, and that any changes to in-scope algorithms (new models, updated models, deprecated models) have been reflected in the inventory. Annual attestation is a documentation exercise; quarterly attestation forces a live catalog that is kept current as a normal part of release management.

The random audit mechanism allows ONC to sample certified installations and verify that the DSI metadata visible inside the live installation matches the attested inventory. This is the mechanism that most changes the operational cost of compliance. The vendor’s attested inventory is one thing; what customers can actually see inside their installation is another thing; the delta between the two is now auditable. If a health system has configured a locally-authored predictive model inside the EHR environment, that model may now be an in-scope DSI for the vendor’s attestation. The vendor’s compliance surface area reaches past what the vendor ships into what the customer configures.

The public posting requirement means the DSI inventory for each certified installation will be published on an ONC-hosted registry, searchable by vendor and by site. The transparency effect is significant. Competitors will see what deployments each vendor is maintaining. Journalists will see which health systems are running which models for which use cases. Plaintiffs’ attorneys will have a searchable inventory of in-use algorithms when preparing litigation. Patients and advocacy organizations will see the AI operating inside their care.

Vendor playbook for Q4 2026 and Q1 2027

Vendors have roughly 10 weeks between the enforcement framework’s publication and the January 1, 2027 effective date. Four workstreams are time-sensitive.

The first workstream is the inventory itself. Every in-scope DSI shipped in the certified product needs to be identified, catalogued, and have complete source-attribute metadata. For a vendor with a long product history, this is a non-trivial lift — there are often dozens or hundreds of embedded predictive algorithms, decision rules, and AI-enabled features that have accumulated over product generations. The inventory needs an owner, a schema, a change-management workflow, and a mechanism for integrating new work going forward. Vendors that have been treating this as a documentation project will find that most of the Q4 2026 engineering budget needs to move onto the inventory work.

The second workstream is the deployment-side metadata exposure. The random audit mechanism means that the DSI metadata must be visible inside the live installation — not just in a vendor-controlled compliance system. The practical requirement is a metadata API inside the EHR that exposes the current in-scope DSI list, their source attributes, and the local configuration state (which DSIs are enabled, which thresholds are configured, which populations are affected). Vendors that have not architected this metadata surface need to design and ship it before January.

The third workstream is customer-side configuration governance. Many vendors allow or encourage customer-side configuration of DSIs — custom predictive models, locally-tuned thresholds, site-specific rules. The enforcement framework treats this configurable surface as part of the vendor’s attestation surface. The vendor needs a governance mechanism that either constrains customer-side configuration to pre-attested parameter spaces, or treats customer configurations as new DSIs that require inclusion in the inventory. Both approaches have significant customer-communication implications, and vendors should pick a direction and begin the customer conversations now.

The fourth workstream is the quarterly attestation workflow. The attestation itself is signed by a named corporate officer. The workflow that produces the attestation needs to pull from the inventory, verify that the source-attribute metadata is complete, verify that deployed DSIs match the inventory, and surface any discrepancies for resolution before sign-off. This is standard governance work, but it is work that most vendors have not had to do before, and it needs to be stood up as a repeatable quarterly process.

Health-system playbook for Q4 2026 and Q1 2027

Health systems have been less in the HTI-2 crosshairs than vendors, but the enforcement framework changes that in ways that are worth being explicit about.

The random audit will reach into the health system’s installation. The audit protocol is published; it covers both vendor-attested documentation and the live deployment state inside the installation. Health systems that have configured custom predictive models, locally-tuned thresholds, or site-specific rules will have those configurations examined as part of the audit. The health system should expect to be a participant in the audit, not an observer.

The public posting will expose the health system’s AI inventory. The DSI inventory for each installation will be published. Health systems that are not currently prepared to have their AI inventory public — for competitive reasons, for privacy reasons, for communication reasons with their clinical staff or patients — need to work through the implications now. The transparency is not optional, and the inventory being made public under HTI-2 compliance will attract press coverage.

The governance capacity gap is the strategic risk. The HIMSS 2026 benchmark data showed that governance maturity lags execution maturity across most U.S. health systems. HTI-2 enforcement is one of the first federal mechanisms that forces a governance-maturity gap to close in a specific time window. Health systems that are at the lower end of the governance maturity distribution will find the compliance lift substantial. Those that have invested in enterprise AI governance over the last two years will find the compliance lift incremental.

Three near-term actions matter. First, build or formalize the enterprise AI inventory. Second, align the inventory with the vendor’s HTI-2 inventory so that the vendor’s attestation and the health system’s local inventory match. Third, stand up a configuration governance workflow that constrains site-level AI configuration to the parameter space that has been pre-attested.

The downstream effects worth watching

HTI-2 enforcement will have second-order effects that extend beyond the compliance workload itself.

The public DSI registry will be used by regulators outside ONC. The FDA will reference it to identify in-use AI/ML devices and verify indication-boundary compliance. CMS will reference it in the context of the Medicare Advantage AI prior-authorization rule. State attorneys general will reference it when investigating utilization-management AI. The registry becomes a shared data layer across multiple regulatory bodies.

The procurement conversation will change. Health systems negotiating new AI purchases will reference the registry to evaluate how vendors have historically specified their DSIs — the completeness of the source-attribute metadata, the clarity of indication boundaries, the subpopulation performance characteristics. Vendors with clean, substantive registry entries will have a procurement advantage. Vendors with thin or vague entries will have procurement friction.

The research literature will accelerate. Academic researchers studying real-world AI performance in healthcare have historically been constrained by the opacity of what is actually deployed. The registry gives the field a sampling frame — a known universe of in-use algorithms, with named vendors, named sites, and documented source attributes. Expect research productivity in post-deployment AI evaluation to increase substantially starting in 2027.

The litigation and insurance landscape will shift. Plaintiffs will use the registry to identify algorithms involved in harm cases. Medical-malpractice insurers will use it to inform premium-setting for health systems running identified high-risk algorithms. Vendors’ errors-and-omissions insurance will become more sensitive to the completeness and accuracy of registry entries.

What good looks like

A well-prepared vendor in Q1 2027 has a complete inventory, operational metadata APIs inside the certified product, a constrained configuration governance workflow with health-system customers, a quarterly attestation process with named ownership, and clean registry entries that read well to a competitor, a regulator, or a journalist. A well-prepared health system has a current enterprise AI inventory, a vendor-aligned HTI-2 inventory, a configuration governance workflow, and a communications posture ready for the registry publication.

Few organizations are at this state today. The organizations that get there by January will be at the leading edge of what AI governance in healthcare looks like under enforceable federal transparency rules. Those that don’t will spend 2027 catching up under enforcement pressure, which is a materially more expensive way to arrive at the same operational state.