Recruiters and HR teams overseeing AI agents face Article 14 of the EU AI Act more often than most, because the Act treats many uses in recruitment and worker management as high-risk. People overseeing those agents must understand them, resist automation bias, interpret their output and be able to override or stop them. Employers must train them and evidence it.
What does Article 14 require in HR?
Article 14 requires that high-risk AI systems can be overseen effectively by natural persons during use. The people overseeing them must be able to understand the system's capacities and limitations, monitor it for anomalies, remain aware of automation bias, interpret its output correctly, decide not to use it or to override or reverse it, and intervene or stop it. Oversight must match the risk, autonomy and context.
The Act treats as high-risk, in substance, AI used to recruit or select people, including targeting job adverts, filtering applications and evaluating candidates, and AI used for decisions on promotion, termination, task allocation based on behaviour or traits, and monitoring or evaluating performance. An HR agent doing any of that is a high-risk system for that purpose.
Article 26 adds deployer duties that bite in HR: assign oversight to competent, trained people with authority, and inform workers' representatives and affected workers before putting a high-risk system into use at work. The GDPR applies throughout, including Article 22 on solely automated decisions and Article 35 on impact assessments.
Who needs oversight competence?
- Recruiters who review shortlists, screening outcomes and rejection drafts.
- Hiring managers who receive agent-prepared candidate summaries.
- HR business partners who use agent analysis in promotion, performance or redundancy processes.
- HR operations who run the agent and set its bounds.
- Works councils or staff representatives, who are not overseers but must be informed and need enough literacy to engage.
Employers in regulated sectors face this twice: once for their HR agents, and again for the credentials of staff who oversee agents elsewhere. Checkable credentials as the new hiring signal for AI operations covers the second side.
Where does an HR agent need a person?
A governed recruitment agent might read applications and job requirements at depth 1, draft shortlists, candidate summaries and rejection messages at depth 2, and schedule interviews within rules at depth 3. Rejections, offers and anything affecting an existing employee's role or pay are held for a named person. In HR, the case for holding is strong even where the platform would allow more: the effects on people are significant and hard to reverse, and the law expects real human involvement.
Overseers need four practical skills.
Spotting proxy criteria. An agent may learn that gaps in employment history, certain postcodes or particular universities correlate with past hiring decisions. These can stand in for protected characteristics. Recruiters must recognise such reasoning in the agent's summary and refuse to act on it.
Reading the facts behind a summary. A candidate summary should show where each claim came from. A qualification confirmed by the issuing institution is signed; one stated on a CV is declared; a claim found on a web page is external. Treating declared and external claims as verified is a common error that training can fix.
Writing a real reason. When a recruiter rejects a candidate the agent ranked highly, or advances one it ranked low, they record why. That record shows oversight took place, and helps if a candidate asks how the decision was made.
Knowing when to stop. If shortlists suddenly skew, for example after a change to a job description, the overseer should be able to pause the agent and return to manual screening.
Competence, authority and intervention in HR
Competence is trained on the organisation's own roles and candidates, with assessment on real past cases, including ones where the agent was wrong.
Authority means a recruiter can advance or reject against the agent's ranking without justification to anyone but the record, and hiring managers are told that the agent's summary is an input, not a verdict.
Intervention means named HR staff can suspend the agent on a requisition or across all hiring. Because an agent's certificate lapses after 90 days, and a run that clears no depth drops it to depth 1 the same day, HR must also be able to screen manually at short notice.
How do you meet Article 14 in hiring and HR?
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Overseers understand limits | Training on the agent's criteria, depths and known bias risks | Training and assessment records per recruiter |
| Awareness of automation bias | Recruiters review a sample of agent rejections, not only shortlists | Sample results and reversals |
| Correct interpretation | Summaries show provenance; recruiters trained to check it | Decision notes citing facts |
| Power to override | Recruiters can advance or reject against ranking within mandate | Overrides with reasons in the record |
| Ability to stop | Named HR staff can suspend the agent per requisition or globally | Suspension events and triggers |
| Competent people assigned (Article 26) | Approval rights tied to current credentials | Credential register matched to decisions |
| Inform workers' representatives (Article 26) | Consultation before go-live, including the training plan | Consultation records |
What will a supervisor, regulator or tribunal ask?
A market surveillance authority, a data protection authority or an employment tribunal may each look at the same decision. Expect: was this a high-risk use, and how did you assess it? Who reviewed this candidate's rejection, and what did they see? How were they trained? Did anyone ever override the agent? Were workers' representatives informed? What did the impact assessment say about the human review step, and is it accurate?
What does Article 14 not cover in HR?
It does not set anti-discrimination law, which applies independently and often more strictly. It does not decide employment law questions, such as consultation requirements in a particular country. It does not set training content. It does not apply to HR agents outside the high-risk uses, such as one answering staff questions about holiday policy. And it does not move responsibility to the provider. Our guide to AI literacy and certification for people who work with agents covers role-based training; related sectors are in Article 14 oversight in credit decisions and what human oversight is.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Where the Academy fits
Pryme Intelligence Academy offers a free readiness assessment for teams of up to 25 people: the team takes the Foundations test set in its industry, and the report shows scores by concept, the gaps, and the first agent to start with. For an HR team about to oversee a recruitment agent, that report shows where training should start before go-live.
Questions people ask
Is an agent that only schedules interviews high-risk?
Scheduling alone is unlikely to be recruitment or selection in the Act's sense. If the agent also decides who gets an interview, it is filtering candidates, and the answer changes.
Do hiring managers need the same training as recruiters?
They need enough to read an agent summary critically and understand it is an input. Recruiters who decide held rejections need deeper training and assessment.
Can candidates ask for a human decision?
Where a decision is based solely on automated processing with significant effect, GDPR Article 22 gives them rights including human intervention. Good practice is to make human review available on request in any case.