🧠 AI Ed Wiki

📄 Research Article

The Paternalistic Filter: Epistemic Injustice and Differential Refusal in LLM-Mediated History Education for Marginalized Romanian Students

● high · 2026-07-14llm, student-experience, equity, higher-ed, bias-mitigation

A systematic API audit of four LLMs acting as history tutors evaluates 1,800 responses about the 1989 Romanian Revolution, exposing a 'paternalistic filter': models differentially refuse or soften answers for marginalized students, reproducing epistemic injustice. The audit reveals that guardrails and refusals are not uniform but patterned by student identity and topic sensitivity.

Centers on LLM tutors in Student Experience and the harms catalogued under AI Tutor Safety Harms, with strong links to Bias Mitigation and Equity in Higher Ed. It extends Over Reliance concerns by showing even refusal behavior can entrench inequality, and calls for fairness evaluations that account for differential treatment across student groups.

Connected Concepts

  • LLM
  • Student Experience
  • Bias Mitigation
  • Equity
  • Higher Ed
  • Over Reliance

    • Connected Articles

  • AI Tutor Safety Harms

    • Citation

    Alexis Popovici, Andrei Ionascu, Adrian-Marius Dumitran (2026). The Paternalistic Filter: Epistemic Injustice and Differential Refusal in LLM-Mediated History Education for Marginalized Romanian Students. arXiv:2607.11292. arXiv preprint.