Account

What new tool do you want?

Legal & Compliance

EU AI Act Risk Checker

Find out which EU AI Act risk category your AI system falls into — prohibited, high-risk, limited risk (transparency) or minimal risk — plus general-purpose AI model duties, the obligations that apply and the key deadlines.

For general information only — not legal advice. Laws vary by location; consult a qualified lawyer.

How to use

  1. 1Answer the questions about what your AI system does and where it's used.
  2. 2See its risk category and the main obligations.
  3. 3Use the result as a starting point for a full legal assessment.

Frequently asked questions

When do the rules apply?

Prohibited practices and AI literacy duties since 2 February 2025; general-purpose AI model rules since 2 August 2025; most other rules, including Annex III high-risk systems and transparency duties, from 2 August 2026, and high-risk AI in regulated products from 2 August 2027. The EU's 2025 'Digital Omnibus' proposal would push some high-risk deadlines back — check the current status.

Does the AI Act apply to companies outside the EU?

Yes, if the AI system is placed on the EU market or its output is used in the EU.

What are the penalties?

Up to €35 million or 7% of worldwide turnover for prohibited practices, and up to €15 million or 3% for most other breaches.

Sources

Why the AI Act sorts systems into tiers instead of one rulebook

The Act doesn't regulate "AI" as a single thing — it matches obligations to how much harm a specific use could cause. A handful of practices (social scoring, manipulative systems, untargeted face-scraping, workplace emotion recognition) are banned outright because no safeguard was judged to make them acceptable. Uses in Annex III areas — employment decisions, credit scoring, education, critical infrastructure, law enforcement — are "high-risk" because they affect access to opportunities or safety, so they carry the heaviest documentation and oversight duties. Systems that merely talk to people or generate synthetic content only owe transparency. The checker's four-tier output follows that same escalation.

Worked example: EU AI Act Risk Checker

Say you select 'Employment — recruiting, CV screening, promotion, firing' for a resume-ranking tool and leave every other toggle at default. The checker returns 'High-risk' immediately — the area alone decides it — with obligations including a risk management system, data governance for training data, technical documentation, human oversight and conformity assessment, applying from 2 August 2026 with a maximum penalty of €15 million or 3% of turnover. Change the area to 'None of these' and switch on only the chatbot toggle, and the same checker now returns 'Limited risk (transparency duties)' with a single obligation: tell people they're interacting with an AI system.

What changes which bucket you land in

Two toggles do the real work: the area selector and the product-safety-component toggle immediately push a system to high-risk regardless of anything else, because Annex III and product-safety law key off the use case, not the technology. The chatbot and synthetic-content toggles only add transparency duties — they don't escalate risk on their own. That's why a marketing chatbot is limited-risk but the same chatbot used to screen job applicants is high-risk: the use case changed, not the model.

Where the classification stops

Annex III allows a narrow carve-out — a system can avoid high-risk status if it performs only a narrow procedural task, improves a completed human activity, or is purely preparatory, but never if it profiles natural persons. Deciding whether your system qualifies requires a documented assessment this tool doesn't perform. The deadlines shown are also in motion: the EU's 2025 'Digital Omnibus' proposal would push some high-risk compliance dates back, so confirm the current timeline before relying on any date here. This is a starting classification, not a compliance determination.

Related tools

More: Legal & Compliance