Compliance Guide for Healthcare

    HIPAA Security Rule Controls for n8n AI Workflows: A Requirement-to-Control Map

    As of November 2026, the HIPAA Security Rule is the operative baseline for n8n AI workflows that handle PHI. HTI-5 is a final rule that streamlines ONC's certification program for health IT developers. It does not set workflow controls for providers. Map each 164.312 safeguard to a control, a location and an evidence artifact, and treat the execution path at the model and tool call as the place where evidence is produced.

    Criteria for assessing control and evidence coverage

    • Can you produce, for any call, a record of what the model or tool was permitted to do and what it did?
    • Is each control located deliberately: platform, infrastructure or execution path, with no row left to policy statements alone?
    • Is the policy enforced at runtime, rather than reviewed after the fact?
    • Does every outbound call that can carry PHI have documented transmission protections and an approved destination?
    • Is there a written determination for the addressable integrity specification in each workflow?
    • Is there a current inventory of workflows, model endpoints, tools and data stores, and a recorded decision on HTI-5 applicability?

    What changed: HTI-5 is a certification rule, not a workflow control source

    Many teams building a compliance map for AI automations expect HTI-5 to supply requirements that healthcare organizations must satisfy in their own workflows. That premise no longer holds. HTI-5, ONC's deregulatory rulemaking, became a final rule in October 2026. It streamlines the ONC Health IT Certification Program for health IT developers. ONC states that of 60 certification criteria, 14 were removed, seven were revised and 39 were retained unchanged. It also scaled back the Real World Testing and Maintenance of Certification requirements.

    Effective dates vary by criterion. Removals of 170.315(a)(12), (c)(4), (e)(3), (g)(1) and (g)(2) take effect January 1, 2027. The revision of 170.315(b)(1) takes effect January 1, 2030. Clinical decision support 170.315(a)(9) is removed and decision support interventions 170.315(b)(11) is revised as of the final rule's effective date.

    The practical consequence is that HTI-5 does not give you technical controls for an n8n workflow. Its relevance is conditional. If your workflows depend on certified health IT that is affected by a removal or revision, track the criterion dates. Document who bears the obligation, the developer or your organization, and do not cite HTI-5 as the basis for a workflow control. The Security Rule carries that load.

    Reading the Security Rule for AI workflows

    The Security Rule sets national standards for electronic PHI created, received, used or maintained by a covered entity or business associate. It requires administrative, physical and technical safeguards for confidentiality, integrity and availability. The sections that matter most for workflow automation are 164.308 (administrative), 164.310 (physical) and 164.312 (technical).

    The text is technology-neutral. The audit controls standard in 164.312(b) requires hardware, software and/or procedural mechanisms that record and examine activity in information systems that contain or use ePHI. It does not name AI, prescribe log fields or mention model calls. Applying it to model and tool calls is your organization's interpretation, and you should record that interpretation. The same applies to transmission security: calls to external model APIs and tool endpoints are transmissions of ePHI over an electronic communications network when they carry PHI.

    HHS's proposed Security Rule update, still a proposal as of November 2026, supports treating AI as in scope. It states that ePHI in AI training data, prediction models and algorithm data maintained for covered functions is protected by the HIPAA Rules. It would also add a written technology asset inventory and network map at 164.308(a)(1)(i). That is not current law, but it is a sensible design target for an inventory of workflows, model endpoints and tool integrations. For self-hosted deployments, 164.310 applies to the facilities and hardware underneath, which the infrastructure layer inherits.

    Requirement-to-control matrix by HIPAA citation

    Each row names the requirement, the workflow-level risk, the control, where it sits and the evidence it should produce. Control location is a design decision for your deployment, and the same requirement often needs controls in more than one place.

    • 164.312(b) Audit controls: Risk: model and tool calls touch PHI with no record of what was permitted or done. Control: record and examine activity at each point where PHI enters or leaves a workflow. Location: execution path, supported by infrastructure retention. Evidence: per-call records of caller, tool or model, policy applied, decision (allow, redact, block, escalate) and time, retained and reviewable.
    • 164.312 Transmission security: Risk: PHI sent to model providers or external APIs over unprotected or unapproved paths. Control: protected transport and approved destinations. Location: infrastructure and execution path. Evidence: egress and network configuration, approved endpoint list, per-call destination records.
    • 164.312 Integrity controls (addressable): Risk: transmitted PHI modified without detection. Control: integrity verification, or an alternative with documented reasoning. Location: infrastructure and workflow design. Evidence: a written determination for each workflow and records of verification results.
    • 164.312 Access control: Risk: an agent or workflow holds broader tool and data access than its task needs. Control: least-privilege credentials, scoped tool permissions, rules on which calls may carry PHI. Location: workflow platform and execution path. Evidence: permission inventory, policy definitions, denial and escalation records.
    • 164.312 Person or entity authentication: Risk: calls cannot be tied to a verified user or service identity. Control: authenticated human and service identities on every workflow and model call. Location: platform and infrastructure. Evidence: identity attribution on each execution record.
    • 164.308(a)(1)(i) Security management: Risk: unknown workflows, models and integrations touching PHI. Control: current inventory of workflow components, model endpoints, tool integrations and network paths, feeding risk analysis. Location: organizational process. Evidence: dated inventory and risk analysis records, with the proposed asset inventory and network map as a design target.

    Platform controls versus execution-path controls

    Workflow platforms generally supply the building blocks: credential storage, user roles, execution history and deployment options that support network isolation. Confirm what your own deployment provides for each, because coverage differs by configuration. Those controls apply to the workflow as a whole. They do not describe what a model was allowed to do with a specific prompt, or whether a specific tool call carried PHI it should not have.

    That gap is where execution-path controls sit. At the model and tool-call level, the control decisions are allow, redact, block, rate limit or escalate for human approval. A prompt containing identifiers can be redacted before it reaches an external model. A tool call outside a workflow's purpose can be blocked. A high-risk action can be routed to a person for approval. Each decision is made at runtime and produces a record as a side effect, which is the opposite of assembling evidence from exports at audit time.

    This is where Trussed fits. Trussed enforces policy in the execution path across models, agents, tools and MCP servers, scoped to individual interactions and tool calls, and generates audit-ready evidence mapped to HIPAA. It runs in production with healthcare organizations and can be self-managed in the customer cloud with configurable retention and data residency. It does not trace whole agentic workflows end to end, so the workflow inventory and platform controls above remain yours to maintain.

    Practical next steps

    Start with the inventory. List every AI-enabled workflow that touches PHI, including models, tools, external APIs and data stores. Build the matrix per citation using the columns above and mark every row where evidence is missing. Record your interpretation of how 164.312 applies to model and tool calls, citing HHS's stated position that ePHI in AI models and related data is protected.

    Then close the gaps in order of exposure. Rows that depend on after-the-fact reconstruction, usually audit controls and access control at the tool-call level, are the first candidates for execution-path enforcement. Rows that depend on documented determinations, such as integrity, need written decisions rather than new tooling. Finally, monitor the proposed Security Rule update, since finalization would change the security management standard, and revisit the HTI-5 criterion dates only if your certified health IT is affected.

    Requirement-to-control test for each AI workflow

    Requirement

    The HIPAA citation or standard that applies to the workflow.

    Control

    The technical measure that satisfies it, and whether it allows, redacts, blocks or escalates.

    Location

    Workflow platform, infrastructure or execution path.

    Evidence

    The artifact the control generates continuously, not after the fact.

    See what your AI workflows can prove at audit

    Walk through the matrix against your own deployment and see how enforcement at the model and tool-call level produces audit-ready evidence mapped to HIPAA.

    Talk to an Expert