AI Resume Screening Audit Reveals Hidden Bias and Incompetence
A recent study released on arXiv (2507.11548) examines eight popular AI platforms utilized for resume evaluation, questioning the belief that these technologies mitigate bias. The research presents the Illusion of Neutrality concept, suggesting that systems seem unbiased primarily due to their inability to effectively distinguish between candidates. In Experiment 1, matched fictitious resumes were employed to assess racial and gender biases, revealing that bias remains in context-sensitive and intersectional ways. Certain models penalize candidates based on demographic indicators, while others demonstrate inconsistent behaviors across various roles and identities. The authors contend that relying solely on fairness metrics is inadequate; evaluating competence is also essential to confirm that AI systems can carry out assessment tasks effectively.
Key facts
- Study audits eight widely used AI platforms for resume screening
- Introduces concept of Illusion of Neutrality
- Experiment 1 evaluates racial and gender bias using matched fictitious resumes
- Bias persists in context-dependent and intersectional forms
- Some models penalize candidates for demographic signals
- Other models exhibit inconsistent patterns across roles and identities
- Fairness metrics alone are insufficient without auditing competence
- Published on arXiv with ID 2507.11548
Entities
Institutions
- arXiv