Small Changes, Big Impact: Demographic Bias in LLM-Based Hiring Through Subtle Sociocultural Markers in Anonymised Resumes

TL;DR

This study reveals that LLMs used in resume screening can infer demographic information from subtle sociocultural markers, leading to biased hiring outcomes even when explicit identifiers are removed.

Contribution

We developed a stress-test framework to evaluate demographic bias in LLM-based hiring, highlighting biases from anonymised resumes with subtle sociocultural cues.

Findings

LLMs can accurately infer ethnicity and gender from anonymised resumes.

Models show systematic bias favoring Chinese and Caucasian males.

Prompting for explanations can increase bias.

Abstract

Large Language Models (LLMs) are increasingly deployed in resume screening pipelines. Although explicit PII (e.g., names) is commonly redacted, resumes typically retain subtle sociocultural markers (languages, co-curricular activities, volunteering, hobbies) that can act as demographic proxies. We introduce a generalisable stress-test framework for hiring fairness instantiated in the Singapore context: 100 neutral job-aligned resumes are augmented into 4100 variants spanning four ethnicities and two genders, differing only in job-irrelevant markers. We evaluate 18 LLMs in two settings: (i) Direct Comparison (1v1) and (ii) Score & Shortlist (Top-Score Rates), each with and without rationale prompting. We find that even without explicit identifiers, models recover demographic attributes with high F1 and exhibit systematic disparities, with models favouring markers associated with Chinese and Caucasian males. Ablations show language markers suffice for inferring ethnicity, while hobbies and activities are utilised for gender. Furthermore, prompting for explanations may paradoxically amplify bias. Our findings suggest that seemingly innocuous markers surviving anonymisation can materially skew automated hiring outcomes.