Deep-Learning Screeners Can Infer — and Penalize — National Origin
National Origin Discrimination in Deep-learning-powered Automated Resume Screening
Sihang Li, Kuangzheng Li, Haibing Lu · arXiv:2307.08624 (PDF) · submitted July 13, 2023
Summary by Andrew Johnson, OneTwo Resume editorial team · updated · We are not affiliated with the authors. arXiv papers are preprints and may not be peer-reviewed.
This pre-LLM study examines deep-learning methods applied to automated resume screening with a focus on national origin — a protected attribute that gets less audit attention than race or gender, despite being richly encoded in resumes through names, schools, languages, and locations.
The authors situate the technical findings in the regulatory gap: laws like GDPR and CCPA gesture at algorithmic fairness, but implementing AI regulation in practice lags the technology, and risks in specific applications go unrecognized.
What the paper reports
- Deep-learning resume screeners can encode national-origin information present in resume text and use it in ways that raise discrimination concerns.
- The paper connects technical audit findings to the difficulty of practically enforcing AI fairness regulation.
What this means for your resume
Our editorial interpretation — the paper does not give job-seeker advice.
- National origin leaks through more channels than a name: foreign universities, languages, visa-status hints, even address formats. You cannot and should not erase your history — but you can avoid volunteering signals that aren’t required, and lead with the experience that’s directly relevant to the role.
Read it with these caveats
arXiv preprint from 2023 studying pre-LLM deep-learning architectures; the specific technical findings predate current systems, though the audit concern transfers directly.
Primary source: National Origin Discrimination in Deep-learning-powered Automated Resume Screening — always read the paper before citing it. Spotted an error in our summary? Tell us and we'll fix it with a visible correction.
More in bias & fairness in ai screening
The Famous Name-Bias Experiment, Rerun on ChatGPT-Era Models
The landmark 2003 field experiment — identical resumes, racially suggestive names — replicated against GPT-3.5, Bard, and Claude.
Resume Search Engines Built on Embeddings Show Race and Gender Bias
Not the chatbots — the embedding models that power "find me candidates like this" search. Audited for gender, race, and intersectional bias.
Newer LLMs Show Less Name Bias — But Judge Your University Instead
A 2025 audit finds explicit gender/race bias has receded in recent LLMs — while implicit bias around educational background remains significant.