+31 (0)88 88 321 88
Home › AI in assessments
AI

AI and assessments

AI now sits inside recruitment: in sorting CVs, in chat conversations and in summarising reports. Below you read what the law watches, which dates apply and which questions to ask a supplier.

An assessment is a measurement. Little has changed there. Scoring a questionnaire happens with a psychometric model and a norm group, not with a language model forming a judgement.

What does change sits around it: pre-selection, interview notes, summaries and chat interviews. That is where AI enters, and where the rules sit. The difference between the two is what a conversation with a supplier should be about.

What the AI Act says

Recruitment counts as high risk

AI systems for recruitment, selection and evaluation sit in Annex III. That calls for documentation, human oversight, logging and information to the candidate.

The date has moved

Those Annex III obligations moved from 2 August 2026 to 2 December 2027 through the digital omnibus. Annex I follows on 2 August 2028.

What does apply already

The Article 50 transparency duty started on 2 August 2026: if someone speaks with an AI system, that has to be clear. Watermarking generated material had a run-in until 2 December 2026.

Banned since February 2025

Emotion recognition at work and in education is not allowed. A system inferring a mood from image or voice falls under that.

Read on

The pages this subject touches within hrmforce.

Keeping it workable

  1. Write down where AI sitsList every place in your process where a model reads or writes along. Without that overview you cannot substantiate the rest.
  2. Keep the decision with a personA model may sort or summarise. The decision to reject someone belongs with a person who can explain it.
  3. Tell the candidate what happensA line in the invitation about what is measured and what happens with the outcome is the minimum. Since August 2026 it is also required for AI conversations.
  4. Keep a record of what you usedWhich version of which system, with which input and which outcome. That is the core of what supervision will ask for.

How hrmforce handles this

The hrmforce questionnaires are scored with psychometric models and norm groups. How heavily an outcome weighs in a decision is something you set in your own procedure.

If you want to know how a score comes about or which data are kept, that is on the scientific accountability and privacy pages. Always ask a supplier for the same: what the model was trained on, who reviews it and how long data are kept.

Read how hrmforce handles data →

Frequently asked questions

May an algorithm decide about a candidate?

The GDPR gives candidates the right not to be subject to a decision based solely on automated processing that has legal effects. A rejection counts as one. So set up your process in a way where a person takes the decision and can explain it, and record which thresholds you use.

Does an online assessment fall under the AI Act?

A classic questionnaire with a fixed scoring key is not an AI system in the sense of the regulation. As soon as a model that learns or decides is added, for instance for pre-selection or a chat interview, you do end up in Annex III.

May you use AI to summarise an interview?

You may, provided you tell the candidate you do it, check the summary before using it and do not keep the recording longer than needed.

Does this apply outside the EU?

The regulation covers systems used in the EU, also when the supplier sits outside it. If you recruit within the EU, it applies.

Questions about AI in your selection process?

An adviser walks through where AI sits and what that means for your procedure.

Free demo