n8n AI Under the EU AI Act: Classify Each Workflow, Not the Tool

n8n is minimal risk as a tool. But that tells you almost nothing — because each AI workflow you build in it has to be classified separately, by what that workflow does.

n8n is a workflow-automation platform with AI capabilities: a node-based visual editor, hundreds of integrations, and a set of AI nodes (built on LangChain) that let you wire LLMs, retrieval, memory and autonomous "agents" into automations. It can be self-hosted. As a piece of automation infrastructure, n8n carries no EU AI Act risk tier of its own. The regulated objects are the AI systems you assemble inside it — and a single n8n instance can run a harmless internal automation alongside one that screens job applicants. The first is minimal; the second is Annex III high-risk.

This page shows how to classify per-workflow, where the Article 25 provider flip catches integrators, and walks a worked example.

---

The Tool Is Minimal; the Workflow Is the Unit of Analysis

Treat n8n the way the Act treats it: as plumbing. Building a workflow that moves data between systems, posts a notification, or summarises an internal document is minimal risk under Regulation (EU) 2024/1689 — no registration, no conformity assessment, no Article 11 documentation. The only standing duty is Article 4 AI literacy (live since 2 February 2025) for the people building and relying on these workflows.

But "n8n is minimal" is not a compliance conclusion. The right question is asked per workflow: what does this automation do, to or for which people? Each AI-enabled workflow is its own AI system, classified by intended purpose under Article 6.

The CV-screening case is the one to watch. It is genuinely easy to build an n8n workflow that pulls inbound CVs, sends them to an LLM node, and ranks or filters candidates. That is AI used for recruitment or selection — Annex III point 4(a) — and the workflow is high-risk, with all the deployer (and possibly provider) duties that follow. The visual simplicity hides the regulatory weight. Conversely, a customer chatbot workflow triggers the lighter Article 50 transparency duty. Run the Article 6(3) filter on any Annex III workflow (narrow procedural task / preparatory work / improving a completed human activity → not high-risk; profiling natural persons → always high-risk), and document it.

---

The Article 25 Flip: When the Integrator Becomes the Provider

Most companies use n8n as a deployer — building internal automations they run themselves, owing the Article 26 duties (use per instructions, human oversight under Article 14 of any output affecting a person, logging, informing workers where relevant).

But Article 25 changes the role if you build a workflow into a product and ship it under your own name or trademark, or substantially modify a system into a high-risk use. An agency that packages an n8n-based screening automation and sells it to clients under its own brand becomes the provider of that high-risk system, inheriting the Article 16 stack: risk management (Art 9), data governance (Art 10), technical documentation (Art 11/Annex IV), human oversight (Art 14), conformity assessment (Art 43), registration (Art 49), Declaration of Conformity (Art 47). The GPAI obligations stay upstream with whichever model vendor powers your AI nodes (Chapter V, Article 53) — your duty is about the workflow-as-system you shipped.

---

The Deadline

If a workflow is stand-alone Annex III high-risk, the obligations apply from 2 December 2027 under the Digital Omnibus deferral, adopted by the European Parliament (16 June 2026) and the Council (29 June 2026) — settled dates, pending only Official Journal publication. Article 50 chatbot transparency is a 2 August 2026 date (not delayed); content-marking lands 2 December 2026. Article 4 literacy and the Article 5 prohibitions — including the emotion-recognition ban — are already in force.

---

Worked Example: Bramble Automations OÜ

Bramble Automations OÜ, an Estonian automation agency with ~25 employees and ~€3M turnover, runs one self-hosted n8n instance with three live workflows:

• An internal "new-deal" notifier that watches the CRM and posts to chat — pure plumbing, minimal risk. Nothing required beyond Article 4 literacy.
• A website support chatbot Bramble built for itself — limited risk, Article 50: it must disclose to visitors that they are talking to an AI system (Art 50(1)) and mark any generated content (Art 50(2)). Applies from 2 August 2026.
• A CV-triage workflow Bramble built for a recruitment client and ships under Bramble's own brand — this pulls applicant CVs into an LLM node and produces a ranked shortlist. That is Annex III point 4(a), high-risk, and because Bramble sells it under its own name, the Article 25 flip makes Bramble the provider (Article 16): full Art 9–15 stack, conformity assessment (Art 43), registration (Art 49). Bramble's lawyer also confirms a fourth, proposed workflow — inferring whether interviewees sounded "confident" from audio — is prohibited under Article 5(1)(f) if run on the client's own staff, and dropped. Deadline for the screening workflow: 2 December 2027 (Digital Omnibus, adopted). SME fine cap under Article 99(6) applies to Bramble.

One instance, four workflows, four different answers — exactly why you classify the workflow, not the tool.

---

How Confir Helps

Register each n8n workflow as its own AI system, and Confir's rule-based engine classifies it by what the workflow does — separating an internal notifier (minimal) from a chatbot (Article 50) from a CV-screening flow (Annex III high-risk) — and derives whether you are deployer or, via Article 25, provider. The engine is deterministic: the same intake yields the same finding, and the rule that fired is written in plain language, so an integrator running dozens of workflows can audit each verdict.

---

Related guides

• recruitment screening AI and Annex III
• Article 50 transparency obligations
• Article 25 value-chain responsibilities
• agentic AI compliance
• provider vs deployer explained
• is my AI high-risk?
• AI system inventory management