The EU AI Act for AI companies
The EU AI Act sets transparency and risk obligations for AI systems sold into the European Union, including disclosure about what a system is and how it was built.
Which edition this covers
Regulation (EU) 2024/1689, consolidated text of 27 July 2026
Checked against the source on 2026-09-13 · read the source
Not covered
- Delegated and implementing acts, and harmonised standards, which are issued separately and are not part of the Regulation text.
- National implementing measures and the penalties each Member State sets.
- Any amendment adopted after the consolidation date above. Check the date before relying on this.
What Trooth witnesses
Read on a schedule, republished with the date
- The transparency disclosures you publish on your Trooth profile and AI system cards
- Your stated risk classification and intended use, republished with the date
- Training-data and model-provenance statements you choose to publish
- Changes to any of the above over time, so drift is visible
What Trooth does not do
Said plainly, so nobody guesses
- Issue a certification, attestation, or audit opinion. Only a licensed auditor can do that.
- Perform a conformity assessment or act as a notified body.
- Decide your risk classification for you. You classify; Trooth records what you published and when.
- Give legal advice or tell you whether you are compliant. That call is yours and your counsel's.
Why this is worth anything to a buyer
A questionnaire answer is a claim typed once and never checked again. EU AI Act here is the rulebook, not the product: what Trooth publishes for it is read from your live systems on a schedule and republished with the date it was last seen, so a buyer checking your company can tell the difference between what was true a year ago and what is true this morning. That is a smaller promise than an audit, and it is one we can actually keep. The full method is on the methodology page.