Conference Paper2026

Class in the Prompt: Language-Based Proxy Bias in AI Systems in Bangladesh and Its Implications for the Global South

Istiaq Ahmed, Faiza Omar Arpita, Ashraful Islam, AKM Mahbubur Rahman, Moinul I Zaber, M Ashraful Amin

Electronic Workshops in Computing

BCS Learning & Development, ISBN: 1477-9358

Abstract

Artificial Intelligence (AI) systems are increasingly used across the Global South for consequential decisions in hiring, credit, and health triage, often before regulatory oversight is in place. This study examines a bias that standard fairness audits are not designed to detect. Large language models (LLMs) produce different outputs for identical content depending on how it is written, without naming writing style in their reasoning; where the language a person writes in signals class position, this becomes a hidden channel of discrimination. We report a matched-guise prompt audit across four frontier LLMs and three decision tasks, varying the register of 30 matched scenarios across five language varieties while holding content constant. In Bangladesh, English from Bangla-medium schooling is scored consistently lower than identical English-medium English across all three tasks and three of four models (gradient: 0.37 hiring, 0.22 loan, 0.17 triage), an effect absent from stated reasoning in 96.4% of cases. We propose a register parity check as a low-cost intervention applicable across the Global South. The findings connect to United Nations (UN) Sustainable Development Goals 10 (Reduced Inequalities) and 16 (Peace, Justice and Strong Institutions), and the governance responses proposed require no new legislation.

Keywords

  • register
  • class position
  • large language models
  • fairness auditing
  • hiring decisions
  • credit decision-making
  • Global South
  • AI governance
  • Bangladesh

CCDS Authors

Istiaq Ahmed
Istiaq Ahmed

Istiaq Ahmed

  • Research Assistant, HCI Wing

Published at

CHI, HCII

Ashraful Islam, PhD
Ashraful Islam

Ashraful Islam, PhD

  • Center Director, CCDS
  • Director, HCI Wing

Published at

CHI, UIST, UbiComp, IJCNN, ICIP, HCII, ICDMW, BLP, ICMLA, Scientific Data, Pattern Recognition, IEEE Internet of Things Journal, JMIR, JMIR Human Factors, JMIR Formative Research, IEEE Access, Smart Agricultural Technology, Diabetes Research and Clinical Practice, PLOS ONE

AKM Mahbubur Rahman, PhD
Dr. Mahbub

AKM Mahbubur Rahman, PhD

  • Director, DS & IPT Wings
  • Associate Professor

    Department of Computer Science and Engineering

    Independent University, Bangladesh (IUB)

Published at

ACL, EMNLP, CHI, ECCV, ECAI, WACV, IPSN, PAKDD, LREC, IJCNN, ICPR, ICIP, HCII, BEA, BLP, Scientific Reports, Pattern Recognition, Information Sciences

Moinul I Zaber, PhD
Moinul I Zaber, PhD

Moinul I Zaber, PhD

  • Director, HCI Wing
  • Supervisor, DS Wing
  • Professor

    Department of Computer Science & Engineering

    University of Dhaka, Bangladesh

  • Visiting Academic

    Department of Computer Science and Technology

    University of Cambridge, United Kingdom

  • Research Fellow

    Institute of Development Studies (Sussex), United Kingdom

Published at

ICPR, HCII, JMIR

Prof. M Ashraful Amin, PhD
M Ashraful Amin, PhD

Prof. M Ashraful Amin, PhD

  • Founder Director, CCDS
  • Director, AI/ML, HCI & DS Wings

Published at

ACL, CHI, UbiComp, ECAI, WACV, PAKDD, LREC, IJCNN, ICPR, ICIP, HCII, CLEF, BLP, ICMLA, Scientific Reports, Pattern Recognition, JMIR, IEEE Access, Smart Agricultural Technology, Frontiers in Computational Neuroscience, Diabetes Research and Clinical Practice, PLOS ONE

References

  1. 1.W. E. Lambert, R. C. Hodgson, R. C. Gardner, S. Fillenbaum. (1960). Evaluational reactions to spoken languages.. Journal of Abnormal & Social Psychology, 60(1), 44–51[10.1037/h0044430]
  2. 2.Valentin Hofmann, Pratyusha Ria Kalluri, Dan Jurafsky, Sharese King. (2024). AI generates covertly racist decisions about people based on their dialect. Nature, 633(8028), 147–154[10.1038/s41586-024-07856-5]
  3. 3.Mehtabul Azam, Aimee Chin, Nishith Prakash. (2012). The Returns to English-Language Skills in India. Economic Development and Cultural Change, 61(2), 335–367[10.1086/668277]
  4. 4.M. Obaidul Hamid, Elizabeth J. Erling. (2015). English-in-Education Policy and Planning in Bangladesh: A Critical Examination. Language policy, 25–48[10.1007/978-3-319-22464-0_2]
  5. 5.Michelle S. Lam, Mitchell L. Gordon, Danaë Metaxa, Jeffrey T. Hancock, James A. Landay, Michael S. Bernstein. (2022). End-User Audits: A System Empowering Communities to Lead Large-Scale Investigations of Harmful Algorithmic Behavior. Proceedings of the ACM on Human-Computer Interaction, 6(CSCW2), 1–34[10.1145/3555625]
  6. 6.Mina Arzaghi, Florian Carichon, Golnoosh Farnadi. (2024). Understanding Intrinsic Socioeconomic Biases in Large Language Models. Proceedings of the AAAI/ACM Conference on AI Ethics and Society, 7, 49–60[10.1609/aies.v7i1.31616]
  7. 7.James Collins. (2017). Dilemmas of race, register, and inequality in South Africa. Language in Society, 46(1), 39–56[10.1017/s004740451600083x]
  8. 8.Sourojit Ghosh, Kyra Wilson. (2025). Bias Is a Math Problem, AI Bias Is a Technical Problem: 10-Year Literature Review of AI/LLM Bias Research Reveals Narrow [Gender-Centric] Conceptions of ‘Bias’, and Academia-Industry Gap. Proceedings of the AAAI/ACM Conference on AI Ethics and Society, 8(2), 1091–1106[10.1609/aies.v8i2.36613]
  9. 9.Sarah V. Bentley, David Evans, Claire K. Naughtin. (2025). What social stratifications in bias blind spot can tell us about implicit social bias in both LLMs and humans. Scientific Reports, 15(1), 30429[10.1038/s41598-025-14875-3]
  10. 10.Loy Lising, Pam Peters, Adam Smith. (2020). Code-switching in online academic discourse: Resources for Philippine English. English World-Wide A Journal of Varieties of English, 41(2), 131–161[10.1075/eww.00044.lis]
  11. 11.Sharmin Rahman Bipasha, MD Nahidul Islam. (2025). Social Class, Language, and Academic Achievement: Understanding English Performance in the Context of Bangladesh’s Stratification. Journal of English Language Studies, 10(1), 39[10.30870/jels.v10i1.31516]
  12. 12.EDMUND O. BAMIRO. (1991). The social and functional power of Nigerian English. World Englishes, 10(3), 275–286[10.1111/j.1467-971x.1991.tb00161.x]
  13. 13.Arcuschin, Iván, Chanin, David, Garriga-Alonso, Adrià, Camburu, Oana-Maria. (2026). Biases in the Blind Spot: Detecting What LLMs Fail to Mention. arXiv (Cornell University)[10.48550/arxiv.2602.10117]