← Back to the homepage

EU AI Act

The AI Act for recruitment firms, clearly explained.

What applies, what is coming, who carries which obligations, and how we build your systems so you are prepared.

Timeline

  1. Aug 2024AI Act in forceThe basis; obligations apply in stages
  2. Feb 2025Ban on certain AI practicese.g. emotion recognition at work
  3. Aug 2026Transparency obligations and oversightThe Federal Network Agency supervises. Originally also the start of high-risk obligations.

What is affected?

High-risk means AI that searches, filters, assesses or shortlists applicants, even if a human makes the final decision. You receive this classification in writing before every project.

  • Winning new search assignmentsNot high-riskAssesses companies, not applicants.
  • CV parsing, scheduling, profiles, reportingLikely not high-riskPreparation and logistics without assessing people.
  • Candidate matching and shortlistHigh-riskWe build and document it to the requirements that apply from December 2027.

Who does what?

We build the system and hand it over to you. You then operate it under your own name and are usually the provider and deployer under the AI Act. We set out the exact allocation of roles in the contract.

What we take care of

  • Written classification before building
  • Build to the high-risk requirements
  • Testing, including for bias
  • Complete documentation at handover
  • Training for your team

What sits with you

  • Declaration of conformity and registration in the EU database, using our documents
  • Human oversight during operation
  • Informing candidates about the use of AI (template from us)
  • Keeping logs for at least six months

Our working principles

  • Classify before buildingEvery project gets a written classification up front. When in doubt, we treat it as high-risk.
  • Humans decideYour team approves every selection that affects people. Messages are created as drafts.
  • Explain, don’t scoreFor each criterion you see what fits and where the information comes from. No opaque scores.
  • No protected attributesAge, gender, origin, religion, marital status and photos are never used in assessments.
  • Everything is loggedInputs, outputs, reasoning and approvals are stored.
  • Tested before handoverWith documented test cases, including checks for unequal results across groups.
  • Clean data, in the EULicensed data sources only, no scraping. Hosted on EU servers; our team accesses data only under a data processing agreement.
  • Kill switchYou can stop every workflow in a single step.
  • Yours to keepAfter handover, the workflows, accounts and documentation are yours. We remove our access.

Your documentation package at handover

So you can meet your obligations as provider and deployer.

  1. Classification
  2. System description
  3. Instructions for use
  4. Data sheet
  5. Risk register
  6. Test report
  7. Logging concept
  8. Change log
  9. Monitoring plan
  10. Training record
  11. Candidate notice (template)

This page is an overview, not legal advice. As of October 2026. We recommend a legal review of your individual case.

Questions about the AI Act in your process?

Book an intro call
Intro call