arXiv:2606.28978v1 Announce Type: new Abstract: We audit fourteen mainstream large language models (LLMs) for hiring discrimination using the paired-resume methodology of Kline, Rose, and Walters (2022). The sole 2023-vintage model reproduces the pro-White callback gap documented in field experiments on labor market discrimination ($+2.12$ pp, significant at the 1\% level). Every model released in 2024 or after shows either a null gap or a significant pro-Black reversal (up to $-3.01$ pp). The same pattern holds on the gender axis. Based on 24,024 paired postings per model across 14 models, ou
The rapid deployment and increasing sophistication of large language models in professional applications, especially hiring, necessitate timely audits of their inherent biases.
This research provides critical empirical evidence on the evolution of bias mitigation in LLMs, indicating a significant improvement in newer models regarding racial and gender impartiality in sensitive processes like resume screening.
The findings suggest that newer LLMs (2024+) are actively being developed with inherent bias reduction, reversing previous discriminatory patterns and potentially setting a new standard for fair AI deployment in hiring.
- · AI developers focused on ethics
- · Individuals from minority groups
- · Companies seeking diverse talent
- · HR tech platforms leveraging advanced LLMs
- · Legacy AI models with unmitigated biases
- · Organizations relying on biased AI systems
- · Developers ignoring ethical AI development
Newer LLMs demonstrate capability for reduced bias in hiring, contrasting sharply with earlier versions.
Increased adoption of ethical and bias-mitigated LLMs could lead to more equitable hiring outcomes and compliance standards.
This could set a precedent for regulatory bodies to mandate bias testing and transparency for AI systems used in critical social and economic functions, accelerating trust in AI.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL