Gender, Race, and Intersectional Bias in AI Resume Screening via Language Model Retrieval
webCredibility Rating
High quality. Established institution or organization with editorial oversight and accountability.
Rating inherited from publication venue: Brookings Institution
Relevant to AI safety discussions around real-world deployment harms, fairness evaluation methodologies, and the governance of high-stakes AI systems in employment contexts; useful for policy and responsible deployment sections.
Metadata
Summary
This Brookings Institution study examines how AI-powered resume screening systems using large language models exhibit measurable gender and racial biases, with intersectional effects that compound disadvantages for certain demographic groups. The research demonstrates that retrieval-based LLM hiring tools can systematically rank candidates differently based on protected characteristics, raising concerns about fairness and legal compliance in automated hiring. It calls for greater scrutiny, auditing standards, and governance frameworks for AI deployment in high-stakes employment decisions.
Key Points
- •LLM-based resume screening systems show statistically significant bias against women and racial minorities, with intersectional combinations producing amplified disparate outcomes.
- •Retrieval-augmented AI hiring tools can embed and perpetuate historical hiring biases present in training data without explicit discriminatory intent.
- •Intersectional bias (e.g., Black women vs. white men) is often worse than additive individual biases, highlighting gaps in single-axis fairness testing.
- •The study recommends mandatory algorithmic audits and transparency requirements for AI tools used in employment screening decisions.
- •Findings have direct policy relevance for regulators, employers, and AI developers navigating emerging AI governance frameworks like the EU AI Act.
Cited by 1 page
| Page | Type | Quality |
|---|---|---|
| AI-Driven Institutional Decision Capture | Risk | 73.0 |
Cached Content Preview
Gender, race, and intersectional bias in AI resume screening via language model retrieval | Brookings
Search
Home
Gender, race, and intersectional bias in AI resume screening via language model retrieval
Contact
Contact
Governance Studies Media Office
[email protected]
202.540.7724
Share
Share
Bluesky Streamline Icon: https://streamlinehq.com Bluesky
Search
Sections
Sections
Contact
Contact
Governance Studies Media Office
[email protected]
202.540.7724
Share
Share
Bluesky Streamline Icon: https://streamlinehq.com Bluesky
Subscribe to the Center for Technology Innovation Newsletter
Sign Up
Research
Gender, race, and intersectional bias in AI resume screening via language model retrieval
Kyra Wilson and
Kyra Wilson
Ph.D. Student
- University of Washington
Aylin Caliskan
Aylin Caliskan
Nonresident Fellow
- Governance Studies , Center for Technology Innovation (CTI)
April 25, 2025
Though the use of AI in the hiring process has continued to grow, few laws have been passed that require auditing of these systems to ensure they do not discriminate against some applicants.
In a simulation of resume screening, some systems resulted in significant gender and racial discrimination, especially for Black men.
Increased protections and transparency with these systems could protect against harmful effects, especially with intersectional identities, and empower applicants to act in the event of discrimination.
The seal of the The United States Equal Employment Opportunity Commission (EEOC) is seen at their headquarters in Washington, D.C., U.S., on May 14, 2021. REUTERS/Andrew Kelly
19 min read
Bluesky Streamline Icon: https://streamlinehq.com Bluesky
Print
Sections
Toggle sectio
... (truncated, 30 KB total)aa9bd39c247651f0 | Stable ID: sid_bLG5bUFP7O