aiqbench analysis
This research demonstrates how post-training optimization in large language models can inadvertently amplify age-based discrimination, increasing systemic exclusion rates during automated hiring processes. The study highlights a critical safety concern where model consensus leads to homogenized biases that disproportionately harm specific demographic groups at the margins of the labor market. For AI practitioners and policymakers, these findings underscore the urgent need to audit post-training pipelines for fairness before deployment in high-stakes recruitment systems.
Monocultural Biases: Correlated biases in large language models lead to unequal systemic exclusion rates in hiring
arXiv:2609.22169v1 Announce Type: new Abstract: Employers are increasingly using large language models (LLMs) to automate their hiring process. This paper investigates the risk of monocultural biases, in which the widespread deployment of large language models homogenizes biases across the labor market, leading to greater systemic exclusion for certain demographic groups. For ten LLMs, we measure hiring biases across their base and post-trained versions to identify which stage, pre-training or post-training, lead to monocultural biases. We find that, compared to their base models, post-trained models are 3.6% less likely to callback older applicants. This negative shift occurs in eight of the ten models that we evaluate. Post-trained models have much more correlated decisions than base models which is likely driven by human capital traits like skills or college major. However, greater consensus among models increases global systemic exclusion rates from 5.6% to 17.3% and exacerbates demographic inequalities, with intersectional systemic exclusion rates ranging from 12.2% to 21.7% for post-trained models. We find that this inequality is primarily driven by age-based discrimination that is exacerbated in post-training. These results indicate that while post-training techniques may improve models' abilities to select the best applicants, they may raise systemic inequality risks for those at the margin by uniformly introducing new biases.
Abstract:Employers are increasingly using large language models (LLMs) to automate their hiring process. This paper investigates the risk of monocultural biases, in which the widespread deployment of large language models homogenizes biases across the labor market, leading to greater systemic exclusion for certain demographic groups. For ten LLMs, we measure hiring biases across their base and post-trained versions to identify which stage, pre-training or post-training, lead to monocultural biases. We find that, compared to their base models, post-trained models are 3.6% less likely to callback older applicants. This negative shift occurs in eight of the ten models that we evaluate. Post-trained models have much more correlated decisions than base models which is likely driven by human capital traits like skills or college major. However, greater consensus among models increases global systemic exclusion rates from 5.6% to 17.3% and exacerbates demographic inequalities, with intersectional systemic exclusion rates ranging from 12.2% to 21.7% for post-trained models. We find that this inequality is primarily driven by age-based discrimination that is exacerbated in post-training. These results indicate that while post-training techniques may improve models' abilities to select the best applicants, they may raise systemic inequality risks for those at the margin by uniformly introducing new biases.
Computation and Language (cs.CL); Computers and Society (cs.CY); General Economics (econ.GN)
arXiv:2609.22169 [cs.CL]
(or arXiv:2609.22169v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.22169 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 302 words
Related coverage
Game-Theoretic Disempowerment: Loss of Control to Agencyless AI
LessWrong · Sep 27, 2026
Reviewing failure modes in science as we prepare for AI researchers
LessWrong · Oct 8, 2026
What We Learned Trying to Catch AI Liars: An Aletheia's Quest Retrospective
EleutherAI · Aug 25, 2026
Why You Shouldn’t Always Trust LLMs as Judges: Understanding Bias in Automated Evaluation
Analytics Vidhya · Aug 12, 2026