In brief
In human resources, AI touches decisions about people: candidate screening, evaluation, planning. The EU AI Act classifies candidate screening at the high-risk level. Three properties of the iDIA framework carry most of the weight: Supervised, because an exclusion gets re-read; Accountable, because it gets attributed; Reproducible, because it gets justified after the fact.
Where is AI used in human resources?
Published · Last updated · Wissam Daibess
Three families of use dominate. Recruitment: posting copy, candidate screening, shortlisting, interview summarization. Performance management: evaluation summaries, drafting assistance. Planning: needs forecasting, retention analysis.
Recruitment draws regulatory attention because the decision rules a person out. The other two families process employee personal information, a category Law 25 protects on the same footing as client data.
Unlisted use is the norm rather than the exception here: an assisted performance review rarely leaves a trace in any inventory.
What risks does AI introduce in HR?
The first is indirect discrimination. A system trained on past decisions reproduces the regularities of those decisions, including the ones nobody owns. The risk is not intent: it is silent reproduction.
The second is supervision in appearance only. An automated shortlist a recruiter approves in bulk is not supervised; it is ratified. The distinction gets documented, or it does not hold up against a challenge.
The third is after-the-fact justification. A rejected candidate can request an explanation. Without documentation of the system, its version and the criteria applied, the organization has nothing to produce.
What does regulation say about AI in HR?
The EU AI Act classifies candidate screening among high-risk uses, with the accompanying requirements for risk management, data quality, human oversight and documentation.
In Quebec, Law 25 protects the personal information of employees and candidates, and governs transparency for decisions made without human intervention. Non-discrimination obligations arise under other regimes, and legal counsel carries their interpretation.
identifiable certifies compliance in AI governance and with Law 25, and aligns practices with the AI Act.
How is AI evaluated in an HR function?
The AI Index evaluates systems and agents one by one, across the six properties of the iDIA framework, scored on documented evidence.
In HR, three properties are read first. Supervised: human review of outputs, calibrated to risk, and the training of those who supervise. Accountable: the person named responsible for each system. Reproducible: configuration documentation, and the ability to replay a past decision.
The Responsible AI Practice designation is granted at the threshold, and only there: the Index must reach it and no property may fall below the floor.
Frequently asked questions
Is AI candidate screening a high-risk use?
Yes. The EU AI Act places candidate screening among high-risk uses, with the accompanying requirements for risk management, human oversight and documentation.
Does bulk approval by a recruiter count as supervision?
A shortlist approved in bulk is ratified, not supervised. Supervision gets documented: where the human intervenes, on what risk criterion, and with what training.
Does Law 25 protect candidate data?
The personal information of candidates and employees is protected on the same footing as client data. An AI tool that processes it falls within the law's perimeter.
Which iDIA properties weigh most in HR?
Supervised, Accountable and Reproducible. They answer the three questions a rejected application raises: who reviewed it, who answers for it, and whether the decision can be replayed.