Free check

The AI Act checklist that shows which duties are yours

Answer a few questions to see the duties that apply to you. Each one shows the article it comes from and whether it applies today or from a later date. You also see which duties are not yours at all.

Your answers stay on this page unless you ask us to email you the result. No account needed.

402

days until the rules for high-risk AI systems apply. The date moved in July 2026, so advice written before then gives a date that has already passed.

The same tool, three different answers

Every duty below shows the article it comes from and the date it starts. A ban in force since February 2025 and a duty starting in December 2027 need very different responses.

  • Summarising your own internal documents

    Minimal risk

    Make sure the people using it understand what it does and where it fails

    Article 4 · In force since 2 Feb 2025

    Proof it's done: A dated record of AI training for each person

  • A support chatbot customers talk to

    Limited risk

    Tell the person they are dealing with a machine, before they start

    Article 50(1) · In force since 2 Aug 2026

    Proof it's done: The wording that says it's a machine, and where it appears

  • Drafting articles you publish

    Limited risk

    Label text written by AI that you publish to inform the public

    Article 50(4) · In force since 2 Aug 2026

    Proof it's done: The label as published, and who is responsible for checking the text

  • Shortlisting job applicants

    High risk

    Have someone competent oversee it, keep its logs, and tell staff before you use it on them

    Article 26 · From 2 Dec 2027

    Proof it's done: Who oversees it, the saved logs, and the notice to staff

The risk depends on what you use a tool for, not on the tool itself. The same subscription can come with no extra duties for one use and be high risk for another. That is why your list of AI tools should record what each one is used for.

FourWinds Digital · fourwindsdigital.com

AI Act scope and obligations checklist

Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744

This result was worked out from the answers given on the page. It shows which rules you need to look at. It is not legal advice, and it was up to date on the day it was printed.

This result is worked out from your own answers. It shows which rules you need to look at. It is not legal advice, and it does not confirm that you comply.

What was answered

Optional, and it does not change the result. If you print the result, the name appears on it.

Does anyone use AI tools for the business?
01Does anyone use AI tools for the business? Not answered

Does anyone use AI tools for the business?

For example, staff writing drafts with an AI assistant, a chatbot on your website, or AI built into a process. A yes brings in the duty to help the staff who use AI understand it, as far as you can (Article 4), whatever the tool is used for.

Does anything work out how staff, candidates or students feel from their face, voice, body or the way they type?
02Does anything work out how staff, candidates or students feel from their face, voice, body or the way they type? Not answered

Does anything work out how staff, candidates or students feel from their face, voice, body or the way they type?

For example video interview software that scores confidence or engagement, a tool that scores staff mood from their voice on calls, or a camera that checks attention in a classroom or exam. Scoring the sentiment of written answers or call transcripts is not this. Count tools you bought as well as ones you built.

Does any AI you use or supply do one of these?
03Does any AI you use or supply do one of these: giving people a social score and using it against them in an unrelated setting; scraping facial images from the internet or CCTV to build a database; using biometrics to sort people by race, politics, religion, trade union membership or sexual orientation; predicting crime from personality alone; or generating intimate images of a real person without consent? Not answered

Does any AI you use or supply do one of these?

Giving people a social score and using it against them in an unrelated setting; scraping facial images from the internet or CCTV to build a database; using biometrics to sort people by race, politics, religion, trade union membership or sexual orientation; predicting crime from personality alone; or generating intimate images of a real person without consent.

What is it used to decide?
04What is it used to decide? Not answered

What is it used to decide?

These are the areas the Act treats as high-risk (Annex III). Pick every one that fits. If none fit, choose “None of these”.

Answer the questions and your checklist appears here, sorted into what applies now and what does not.

What this does not do

This checklist shows which duties apply to your business. It is not your compliance record. It does not keep a list of your AI systems, the proof that each duty was met or the dates each one falls due. It is not legal advice either.

For the transparency duties in full (Article 50), including wording you can publish, use the transparency check. To see why the dates on this page fall where they do, read the deadline that moved. That article explains what the 2026 changes to the Act (the Digital Omnibus) moved and what they did not.

The free compliance record

AI Act included.

Keep the free compliance check and your record holds an AI register and an AI literacy record. Both are free.

  • The AI register lists every AI tool you use, what the AI Act asks of each use, and the decisions you record about it.
  • The AI literacy record keeps who was briefed on which tool, what they were shown, and when.

Article 4 asks for measures that support AI literacy, and sets no level any one person must reach. It has applied since 2 February 2025. Duties for high-risk uses apply from 2 December 2027.

Take the free compliance check Talk to us about the record