Free · Powered by Claude AI · 60 seconds

Are you over-classifying your AI?

33% of EU AI startups self-classify as Annex III high-risk. EC's official estimate: 5-15%. A 2-7× over-classification gap (Vestbee survey 2025, n=113) — driven by fear, not analysis. This deep classifier runs all 8 Annex III categories + GPAI Article 53 + Article 50 — and tells you de-classification verdict with sub-paragraph precision. Proving you're NOT high-risk is half the work of an audit.

8 Annex III categories 5 priority actions 0 signup
No data stored Claude AI powered Updated 2026 Built by GRC operator
01 · Company name or domain
02 · Industry closest match
03 · AI features select all that apply
04 · Use case 1-2 sentences
05 · EU exposure
Analyzing your AI deployment
Claude is mapping your stack against EU AI Act articles…
Mapping AI features → Annex III categories
Evaluating GPAI Article 53 applicability
Checking Article 50 transparency obligations
Generating 5 priority remediation actions
How this works: The classifier sends your inputs to Claude (Anthropic's frontier AI) with full EU AI Act context (Articles 6, 50, 53, Annex III). Results are non-legal informational guidance produced by an LLM and reflect a snapshot of the regulation as of May 2026. Edge cases — multi-modal AI, complex provider/deployer structures, biometric inference — benefit from human review. No data is stored unless you opt-in to the audit-readiness PDF. Full Article 11 technical documentation: Model Card.

Common questions

How is this different from the 5-question Quick Check?

The Quick Check is a static decision tree — it gives you a tier (Annex III high-risk, GPAI, Limited Risk, Minimal) in 60 seconds without context. This Deep Classifier sends your specific stack to Claude AI with the full EU AI Act text, returns category-by-category mapping across all 8 Annex III items, and generates 5 priority actions tailored to your industry and AI features. Think of it as the "explain my classification" upgrade.

What data leaves my browser?

Your form inputs (company name, industry, AI features, use case description, EU exposure) are sent to the eucomplyhub API which forwards them to Anthropic's Claude API for classification. We log a notification email to ourselves containing the classification verdict (no further use). We do NOT store the form data in a database. Anthropic's API terms govern their handling — they don't train on API customer data.

Why use Claude AI for compliance classification?

EU AI Act has hundreds of cross-references and category boundaries that aren't obvious from keyword matching. Claude reads the full regulation context (Articles 6, 9-15, 27, 50, 53 + Annex III), applies it to your specific deployment description, and explains its reasoning category-by-category. This catches edge cases (e.g. dual GPAI deployer + Annex III provider classifications) that rule-based tools miss.

How accurate are the priority actions?

The 5 actions are generated by Claude based on your inputs and EU AI Act obligations. They cite specific articles (e.g. "Article 14 human oversight UI", "Article 9 risk management log") so you can verify against the regulation. ~90% of mid-market SaaS scenarios get accurate actions. Edge cases (multi-jurisdiction deployments, GPAI fine-tuning, biometric inference) benefit from a personalized human review — request via the audit CTA below the results.

What if I don't agree with the classification?

Reply directly to the email notification you'll receive after using the tool, or use the "Reply with questions" button on the results screen. The classifier defaults to conservative (lean toward applicability) — over-classifying is safer than under-classifying for compliance. If you have nuance Claude missed, send the details and you'll get a manual re-review within 24h.