EU AI Act High-Risk Screening n8n Workflow w/ Groq & Data Tables

Screen your AI systems for EU AI Act High-Risk signals—automatically in n8n

This n8n workflow gathers AI system details from a form, checks EU AI Act high-risk indicators using configurable data-table keyword matching and a Groq-hosted LLM, then produces a risk tier (including evidence and obligation lists) and logs the result to an internal register table.

What this workflow does

• Collects AI system inputs via an n8n form: purpose, decision effects, whether it affects people, biometric/emotion data usage, human review, role, and EU market scope.
• Loads configurable thresholds and Annex III keywords from an n8n data table (Annex III category keywords and notes).
• Runs deterministic screening rules to detect prohibited-practice patterns (Article 5 signals), match Annex III areas by keywords, and score additional signals such as significant effects on people, biometric/emotion data use, and lack of human review.
• Uses Groq for free-text analysis: sends the user’s free-text purpose to a Groq chat model to suggest relevant Annex III points, determine whether the system profiles natural persons, and return a confidence score (parsed from structured output).
• Combines rule-based + LLM signals into a final risk tier: PROHIBITED, HIGH_RISK, NEEDS_REVIEW, LIMITED_RISK, or MINIMAL_RISK.
• Generates an evidence report with the tier, score, annex points, signal codes, and the obligation list applicable to the outcome.
• Registers the assessment by writing tier/score/report details to an n8n data table and shows a tailored completion page (with different behavior for minimal risk vs. non-minimal risk).

Use cases

• EU SaaS operators and product teams need a repeatable EU AI Act high-risk screening workflow for new AI features.
• Automation engineers want an auditable process for documenting Annex III screening outcomes and risk tiering.
• Teams testing AI governance: capture intent and human-review assumptions, then log decisions to an internal register.

Technical details

• n8n nodes: form, data table, if, set, code, and a sticky note.
• LLM integration: Groq chat model for structured suggestions on Annex III points, natural-person profiling detection, and confidence scoring.
• Configurable setup: create and select an n8n data table for thresholds, Annex III keywords/notes, and the internal register logging.