Self-Hosted n8n vs n8n Cloud for HIPAA: What Changes for PHI Workloads
Neither hosting model makes n8n HIPAA compliant on its own. Hosting changes who operates the runtime, who must sign a BAA, and who produces evidence. It does not change the covered entity's or business associate's Security Rule obligations, and it does not govern what a workflow sends to LLMs, APIs, and tools. Compare the two models by responsibility boundary, then close the remaining gaps with explicit controls.
What a PHI workflow needs before production
- A written BAA for the n8n offering in use, with scope and exclusions recorded.
- A data-path map from trigger to runtime, stored execution data, credentials, and each outbound call, with a responsible party and BAA status per hop.
- A named owner for each 164.312 safeguard in the chosen model, and who evidences it.
- Defined retention and deletion for execution data and logs, with PHI minimized before it is stored.
- A documented decision for each LLM, API, and tool recipient: BAA in place, or reasoning for why one is not required.
- A record of the policy applied to each interaction and tool call, and the outcome, retained as evidence.
The regulatory frame comes before the hosting choice
The Security Rule technical safeguards in 45 CFR 164.312 apply to covered entities and business associates, wherever the software runs. Self-hosting does not remove those obligations, and a hosted service does not transfer them away.
What changes is the cloud relationship. HHS lists a cloud service provider that creates, receives, maintains, or transmits ePHI on a customer's behalf as an example of a business associate. A covered entity or business associate may use a cloud service for ePHI if it has a HIPAA-compliant BAA with the provider and otherwise complies with the HIPAA Rules. HHS also states that using a cloud service provider to maintain ePHI without a BAA is a violation.
HHS adds that cloud configuration, whether public, hybrid, or private, may affect the risk analysis, risk management plans, and BAA provisions of all parties. That is the right way to read this comparison. Each hosting model is an input to your risk analysis, not a compliance status. Whether a hosted n8n Cloud service acts as a business associate depends on whether it creates, receives, maintains, or transmits ePHI for you, and on the terms the vendor will put in writing for your specific offering.
Responsibility boundary, side by side
The table describes where responsibility generally falls by deployment model. The exact split in a hosted model is set by the vendor's contract and BAA, so treat the cloud column as the question to answer in writing, not a given.
- BAA: Self-hosted: no BAA is needed with the vendor for the runtime you operate, but you still need one with every downstream recipient of ePHI. n8n Cloud: a BAA covering the specific offering, features, and data types is required before any ePHI flows.
- Infrastructure hardening: Self-hosted: the customer owns patching, network segmentation, host configuration, and backups. n8n Cloud: the vendor operates the infrastructure, and the contract defines what it commits to.
- Access control: Self-hosted: the customer configures authentication, roles, and administrator access to the instance. n8n Cloud: shared, with the customer still deciding who can build, run, and view workflows.
- Encryption and credential storage: Self-hosted: the customer configures encryption in transit and at rest and manages the keys that protect stored credentials. n8n Cloud: the vendor handles platform-level protections, and the customer decides which credentials to place in the platform.
- Data residency and retention: Self-hosted: location and retention of execution data are set by the customer. n8n Cloud: set by the vendor's offering and terms, so confirm both before use.
- Audit evidence: Self-hosted: the customer builds and retains it. n8n Cloud: HHS notes HIPAA does not require a provider to share security documentation or allow customer audits, so evidence support must be contracted.
PHI exposure points that remain in both models
Hosting decides who runs the engine. It does not decide where PHI travels once a workflow executes. Four exposure points persist in either model.
First, stored execution data. A workflow that processes a patient record can leave inputs and outputs in execution history, so retention settings and deletion control matter in both models. Second, stored credentials. API keys and tokens for clinical systems, model providers, and tools are access paths to PHI and need scoped permissions and rotation. Third, third-party nodes and integrations, each of which is another party that may touch the data. Fourth, outbound calls to LLMs, external APIs, and tools.
The fourth point is the one most often missed. HHS guidance extends BAAs from business associates to their subcontractors. Any model or tool provider that receives ePHI from a workflow raises the same BAA question, and neither n8n deployment model answers it for you. HHS also notes that no BAA is required with an app developer that does not create, receive, maintain, or transmit ePHI for you. That makes the scoping decision a documented judgment for each recipient, and it should be recorded with its reasoning.
Minimizing what reaches each recipient is the most reliable mitigation. A prompt that carries only the fields a step needs creates a smaller disclosure than one carrying the whole record.
Plan for rule changes without treating them as current law
As of November 2026, the 2025 HIPAA Security Rule NPRM is a proposed rule. HHS OCR describes it as the first proposed update to the Security Rule since 2013, and it would require regulated entities and their business associates to strengthen protections for ePHI against external and internal threats. It is not current law, and this comparison does not apply it as such. Architectures that already produce clear control ownership and durable evidence are better placed to adapt if requirements tighten.
The practical consequence is that vendor assurance should come from contract terms and your own controls, not from assumption. A hosting choice that looks simpler on day one can leave evidence gaps that surface later.
Where runtime policy enforcement fits
A runtime enforcement layer is a design option for the gap neither hosting model closes: controlling what a workflow does at the moment it calls a model or tool. Placed in the execution path, it can apply a policy to each interaction and decide whether to allow, redact, or block it, so PHI is handled before it leaves rather than reviewed after the fact.
Trussed AI provides this for regulated enterprises. It enforces policy in the execution path across models, agents, tools, and MCP servers, with actions that include allow, redact, block, rate limit, and escalation for human approval, scoped to individual interactions and tool calls. Enforcement generates audit-ready evidence mapped to HIPAA, showing what was allowed and what happened. It deploys self-managed in the customer cloud, with configurable retention and data residency, which suits teams that self-host their workflow runtime and want policy and evidence to stay inside their own boundary. Trussed runs in production with healthcare and pharma organizations.
This is a complement to either hosting model, not a substitute for a BAA, a risk analysis, or your own safeguards. It does not make n8n HIPAA compliant, and it does not replace the responsibilities described above.
Where responsibility sits for a PHI workflow
Runtime and infrastructure
Operated by the customer when self-hosted. Split with the vendor, by contract, when hosted.
BAA
Required with any cloud service provider that maintains ePHI. Scope must be read, not assumed.
Stored data and credentials
Execution data, logs, and credentials can hold PHI in either model.
Outbound calls
LLM, API, and tool calls are separate disclosure points in either model.
Evidence
What the workflow was allowed to do and what it did must be demonstrable.
Map your PHI workflows to a responsibility boundary
Review how policy enforced in the execution path and audit-ready evidence apply to workflows that call models and tools.
Request a Demo