What HTI-5 Changes for AI Agents Built in n8n at Health Systems
As of November 2026, HTI-5 is a finalized deregulatory rule that streamlines the voluntary ONC Health IT Certification Program, and it is aimed at health IT developers. It removes or revises certification criteria, including clinical decision support and decision support interventions. It does not replace the HIPAA Security Rule duties a health system already has for ePHI, including audit controls under 45 CFR 164.312(b). For n8n agents, the practical work is to determine whether each workflow touches ePHI or certified health IT, then decide which controls belong in workflow design and which belong in the execution path, where individual model and tool calls can be allowed, redacted, blocked or escalated and recorded as evidence.
Where the control points are in an n8n agent
- Trigger and input: classify data and decide what may enter the workflow
- Prompt to model: inspect and redact before the call leaves
- Tool or API call: check the action and parameters against policy
- Tool response: inspect what returns toward the model
- Sensitive actions: route to human approval
- Every decision: record it as evidence
Evaluating runtime enforcement for your n8n deployments
- Does it evaluate each individual model call and tool call in the path, and which actions does it support (allow, redact, block, escalate)?
- Which HTI-5 provisions, if any, does the vendor say apply to your workflows, and where in the final rule text?
- What is recorded per interaction, where is it stored, for how long, and can it be exported?
- Does the evidence store hold PHI, and how is it protected under the Security Rule?
- Can it be deployed in your own cloud, with your retention and data residency settings?
- Is the applicability determination for each workflow (ePHI, certified health IT) documented, with an owner?
What HTI-5 actually changes
HHS announced the HTI-5 final rule on 26 October 2026. Official materials describe it as a deregulatory action that reduces compliance costs for health IT developers by streamlining the ONC Health IT Certification Program, which is voluntary. The changes are written as removals and revisions of criteria in 45 CFR 170.315.
According to the final-rule criteria table, clinical decision support (a)(9) is removed and decision support interventions (b)(11) is revised, both effective on the final rule's effective date. Security tags for summary of care (b)(7) and (b)(8) and care plan (b)(9) are also removed on that date. Family health history (a)(12), clinical quality measures filter (c)(4), patient health information capture (e)(3), automated numerator recording (g)(1) and automated measure calculation (g)(2) are removed effective 1 January 2027. Transitions of care (b)(1) is revised effective 1 January 2030.
These are criteria for certified health IT. They are not written as requirements for a health system's own orchestration layer. Whether any of them bears on an n8n agent depends on facts the health system has to establish, chiefly whether the workflow relies on certified health IT or on EHR APIs from a certified developer. Counsel should review that determination before anyone describes an HTI-5 obligation as applying to the health system.
Three layers to keep separate
Policy documents for agent workflows tend to blur three different sources of requirements. Keeping them distinct makes the applicability analysis auditable and avoids attributing to a rule something it does not say.
- HTI-5: Certification program changes for health IT developers. Relevance to a workflow depends on whether it relies on certified health IT or certified-developer EHR APIs. Cite only the final-rule dates.
- HIPAA Security Rule: Existing duties for covered entities and business associates over ePHI they create, receive, use or maintain. Technical safeguards in 45 CFR 164.312, including audit controls in 164.312(b), apply where an agent workflow handles ePHI.
- Internal practice: Controls that go beyond both, such as per-call policy enforcement and human approval for specific tool actions. These are design choices, not requirements of HTI-5 or HIPAA.
Workflow design versus the execution path
Some controls belong in the workflow itself. Scope each agent to the minimum set of tools, use separate credentials per workflow, avoid passing full records when a few fields suffice, and set retention for execution data deliberately. These choices are made at build time and reviewed with the workflow.
Other controls belong in the execution path, because they depend on what actually happens in a given interaction. A prompt that was safe in testing may carry PHI in production. A model may choose a tool call nobody anticipated. Build-time review cannot settle those cases. Policy has to be enforced at runtime, not reviewed after the fact, with outcomes of allow, redact, block, or escalate for human approval.
Trussed is one way to do this. It enforces policy in the execution path across models, agents, tools and MCP servers, scoped to individual interactions and tool calls, and produces audit-ready evidence from that enforcement. It runs in production with healthcare and pharma organizations and can be self-managed in the customer's cloud with configurable retention and data residency. It does not trace whole agentic workflows end to end or govern at the level of a business transaction, so those needs require other measures.
This placement is an architectural proposal. Neither HTI-5 nor HIPAA mandates it.
What evidence an auditor should be able to see
Audit controls under 164.312(b) concern recording and examining activity in systems that contain or use ePHI. For an agent, that translates into being able to answer two questions for any interaction: what was the agent permitted to do, and what did it do.
A useful per-interaction record includes the policy version applied, the decision taken (allowed, redacted, blocked or escalated), the identity or workflow on whose behalf the call ran, the target model or tool, and a timestamp. For escalations, it should also capture who approved or rejected the action. Evidence is more defensible when it is produced by enforcement itself rather than reconstructed later from scattered execution data.
The evidence store raises its own question. If records contain PHI, the store falls under the Security Rule too, so decide whether to keep content out of evidence, and how access, retention and export are controlled. Evidence should be exportable in a form an auditor can read without database access.
Where controls sit in an n8n agent path
Workflow design
Scoping, credentials, data minimization, applicability decisions
Model and tool calls
Policy enforced per interaction: allow, redact, block, escalate
Evidence
Per-call record of policy, decision, identity and time
Applicability review
ePHI and certified health IT determinations, documented
Map your n8n agents to enforcement and evidence
Walk through one workflow with us: where ePHI flows, which calls need policy enforced in the execution path, and what audit-ready evidence each decision should produce.
Request a DemoHow does your AI governance program compare?
See where your program has gaps in less than 2 minutes.