Healthcare AI Compliance Comparison

    HTI-5 vs HTI-1: Decision Support Source Attributes in n8n Clinical Workflows

    As of November 2026, HTI-5 is final and revises § 170.315(b)(11) to fully remove the source attribute support and source attribute access and modification requirements that HTI-1 introduced. Those requirements bound certified health IT developers, not custom orchestration layers such as n8n. For n8n clinical workflows, the practical effect is that a certified EHR can no longer be assumed to supply source attribute information. Model descriptions and per-call evidence become the organization's responsibility, and HIPAA audit controls apply at the point where a workflow calls a model or tool and touches PHI.

    Where control and evidence attach in an n8n workflow

    1. 1

      1. Trigger and EHR read

      Record which EHR API call retrieved PHI, for which purpose, and which fields were returned.

    2. 2

      2. Pre-model preparation

      Decide which input categories may reach the model. Redact or block fields that are not needed for the intervention.

    3. 3

      3. Model call

      Attach model identity, version and descriptive metadata to the call record, and record the policy decision on the request.

    4. 4

      4. HTTP or tool node

      Treat each outbound call as its own boundary. Capture the destination, the data sent and the policy outcome.

    5. 5

      5. Human approval and write-back

      Where a recommendation affects care, escalate for approval and record who approved before any EHR write.

    Evaluation criteria for runtime enforcement and evidence

    • Enforcement and evidence at each model and tool call that handles PHI, not only at workflow start or end.
    • Per-interaction records: which workflow or agent called which model or tool, with what inputs, under which policy decision.
    • Policy outcomes of allow, redact, block or escalate for human approval, configurable independently of the EHR vendor's certified capabilities.
    • Model and intervention descriptive metadata linked to each call record and producible on request.
    • Configurable retention, protected storage and data residency for audit evidence.
    • Self-managed deployment in your cloud so PHI and evidence remain in your environment.

    What HTI-1 required for source attributes

    HTI-1 was published in the Federal Register on January 9, 2024. It required Health IT Modules certified to § 170.315(b)(11) to support source attributes in (b)(11)(iv)(A) and (B). These are categories of technical performance and quality information for both evidence-based and predictive decision support interventions (DSIs).

    HTI-1 defined four capabilities: access to complete and up-to-date plain language descriptions, an indication when information is not available for a specified attribute, the ability to record source attribute information, and the ability to change it. The criterion required support for the fields, not necessarily responsibility for their content. Developers were not responsible for the accuracy or use of attribute information that users modified.

    Scope was also limited. The access requirement covered evidence-based and predictive DSIs supplied by the developer as part of its Health IT Module, for a limited set of identified users. Access to predictive DSI attributes was required only if the developer supplied the predictive DSI. Developers were not accountable for populating attributes for predictive DSIs they did not supply, including other-party models used within their certified health IT. The Certification Program did not prescribe a display format.

    HTI-1 and HTI-5 side by side

    The table compares the two rules on the points that matter for workflow design. HTI-5 was published in the Federal Register on October 28, 2026. Its changes to § 170.315(b)(11) take effect on the effective date of the final rule.

    • Source attribute support, (b)(11)(iv): HTI-1: certified modules must support the attribute categories for DSIs. HTI-5: requirement fully removed as part of the 'model card' scope reduction.
    • Access and modification, (b)(11)(v): HTI-1: access, record and change capabilities, limited to developer-supplied DSIs. HTI-5: requirement fully removed.
    • Criterion status: HTI-1: (b)(11) introduced. HTI-5: (b)(11) listed as Revise. Clinical decision support, § 170.315(a)(9), is listed as Remove.
    • Who is bound: Both rules operate through the ONC Certification Program on Health IT Modules and their developers. Neither is described as certifying a custom orchestration workflow.
    • Content responsibility: HTI-1: developers supported the fields but were not necessarily responsible for content. HTI-5: no certification-level source attribute fields to support.
    • Rulemaking scale: HTI-5 finalized removal of 14 certification criteria and revision of seven, against a proposal to remove 34 and revise seven.

    Who is responsible for what after HTI-5

    Certification obligations fall on certified health IT developers for their Health IT Modules. HIPAA technical safeguards under 45 CFR 164.312 fall on covered entities and business associates. These are separate questions, and an n8n workflow should not be described as certified or ONC-regulated simply because it calls a model that supports clinical decisions.

    Because HTI-5 removes the criterion-level source attribute requirements, an EHR can no longer be assumed to expose model descriptions, input categories or logic descriptions as a certified capability. If the EHR vendor supplies a predictive DSI, that information may still be available as a product capability, and contract terms are the way to secure it. If the model is called from the orchestration layer, the organization must hold that metadata itself.

    This is an architect's judgment, not a regulatory mandate: dropping source-attribute-style documentation because the certification requirement is gone creates a gap. Whether internal AI governance still requires it is a policy decision. Most clinical teams will want model identity and version, intended use, input categories and a plain description of the logic available on request. Record where each item came from: the vendor, an internal model owner, or a user edit. HTI-1 stated that developers were not responsible for the accuracy of user-modified attributes, and the same provenance question applies to anything your own teams edit.

    How Trussed applies policy in the execution path

    Trussed enforces policy in the execution path across models, agents, tools and MCP servers. Decisions are enforced at runtime, not reviewed after the fact. Scope is the individual interaction or tool call: each can be allowed, redacted, blocked, rate limited or escalated for human approval. Enforcement generates audit-ready evidence mapped to HIPAA, among other frameworks, so a reviewer can see what a model or tool call was permitted to do and what happened.

    Trussed deploys self-managed in the customer cloud, with configurable retention and data residency. It runs in production with healthcare and pharma organizations, holds SOC 2 Type II and ISO 27001 certifications, and supports HIPAA requirements. It does not replace the EHR vendor's product capabilities or your own model documentation, and it does not decide what your governance policy should require. It enforces that policy at the call, and it produces the evidence that policy was applied.

    Where source attribute responsibility sits after HTI-5

    Certified health IT developer

    Certification criterion § 170.315(b)(11) no longer includes source attribute support or access and modification.

    Healthcare organization

    Owns model descriptions, governance decisions and vendor contract terms for AI-enabled steps.

    n8n workflow layer

    Model calls, HTTP or tool nodes and EHR API calls are where PHI crosses boundaries.

    Execution-path enforcement

    Policy decisions and evidence are recorded per interaction and per tool call.

    Review enforcement at the model and tool-call level

    Walk through how policy decisions and audit-ready evidence are produced for individual interactions and tool calls in a self-managed deployment.

    Request a Demo