AI can screen, rank and assess job applicants. But can employers prove that its decisions are fair?
A new review of 40 representative studies and industry systems suggests that safeguards are not developing as quickly as the technology. None of the AI recruitment systems examined jointly evaluated usefulness, fairness, privacy and security.
The findings do not prove that these tools discriminate against women. However, unfair treatment can be difficult to detect when problems arise at several stages of the hiring process.
AI Recruitment Is Moving Beyond CV Screening
The new paper, From Matching Models to Recruiting Agents, traces how AI recruitment has evolved from simple matching tools into systems that can influence several stages of hiring. Its authors reviewed 40 representative academic studies and industry systems, alongside relevant legal and industry sources.
Earlier tools were primarily designed to compare CVs with job descriptions or rank applicants. Newer recruiting agents can perform more complex, connected tasks: finding potential candidates, gathering evidence about their experience, comparing their suitability and supporting—or even executing—the next step in the process.
AI can now be involved in candidate sourcing, ranking, assessment and interviewing, as well as the handoff between automated tools and human recruiters. This means that a single recruitment decision may be shaped by several automated steps, making it harder to see where an error or unfair outcome began.
Read also: Portfolio Career: How to Do More Than One Thing Without Losing Your Mind
The Technology Is Advancing Faster Than Its Safeguards
As AI recruitment systems become more capable, the way they are evaluated may not be keeping pace. None of the systems in the researchers’ coded sample was assessed across four essential areas at once: usefulness, fairness, privacy and security. Privacy was not directly evaluated in any of them.
Looking only at a system’s final recommendation may also hide problems that occurred earlier. A qualified applicant might have received a lower ranking because the system failed to interpret her experience correctly—or she may never have been presented with the opportunity in the first place.
Historical behavioural data add another layer of risk. Past clicks, applications and hiring decisions do not necessarily reveal who was best suited to a role. They may also reflect which candidates saw the vacancy, who felt encouraged to apply and which profiles employers previously preferred.
If AI treats those patterns as evidence of suitability, it may reinforce old hiring habits instead of identifying the strongest candidates.
Why This Matters for Women and Other Underrepresented Candidates
Bias does not arise from data alone. As NIST explains, human and institutional factors can also influence how AI systems are developed and used.
AI recruitment systems learn from information created in the real world, including records of past applications and hiring decisions. If those records reflect existing preferences or unequal access to opportunities, the technology may learn the same patterns rather than remove them.
Read also: 15 Small Workplace Habits That Can Transform Your Career
What Employers Should Ask Before Using AI to Hire
The EU AI Act classifies certain AI systems used in recruitment and worker management as high risk, reflecting how significantly these tools can affect people’s opportunities.
Before introducing AI recruitment tools, employers need to understand more than how much time or money they could save. They should ask:
- Which hiring decisions will the system influence?
- What data was it trained and tested on?
- Has its performance been evaluated across different candidate groups?
- Can applicants request an explanation or challenge a decision?
- At which stages will a human review the system’s work?
- Are privacy and security tested alongside accuracy and performance?
Human oversight must be meaningful. Recruiters need the authority to question and override automated recommendations.
AI can make recruitment faster, but speed alone does not make hiring fair. As these systems gain more influence, employers must ensure that every automated decision remains open to human review.
