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

arXiv: https://arxiv.org/abs/2607.11292 · HTML: https://mlsanity.com/paper/2607.11292/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 7 of its 10 references that are in the map.
- First posted on arXiv 13 Jul 2026.

## Builds on (most cited first, 7 of 7)

- [OR-Bench: An Over-Refusal Benchmark for Large Language Models](https://mlsanity.com/paper/2405.20947/) (2024)
- [Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Models](https://mlsanity.com/paper/2305.18189/) (2023)
- [Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs](https://mlsanity.com/paper/2311.04892/) (2023)
- [LLMs are Biased Teachers: Evaluating LLM Bias in Personalized Education](https://mlsanity.com/paper/2410.14012/) (2024)
- [Epistemic Injustice in Generative AI](https://mlsanity.com/paper/2408.11441/) (2024)
- [A taxonomy of epistemic injustice in the context of AI and the case for generative hermeneutical erasure](https://mlsanity.com/paper/2504.07531/) (2025)
- [Cultural Biases of Large Language Models and Humans in Historical Interpretation](https://mlsanity.com/paper/2504.02572/) (2025)

## Abstract (arXiv)

As Large Language Models (LLMs) are increasingly deployed as conversational tutors, they risk institutionalizing systemic inequalities. This study presents a systematic API audit of four LLMs acting as history tutors, evaluating 1,800 responses regarding the 1989 Romanian Revolution across five student personas varying by ethnicity and socio-economic tier. We uncover four interconnected patterns of epistemic paternalism: (1)~Differential Refusal, where safety-aligned models block 76.7% of educational requests from low-tier students; (2)~Epistemic Gatekeeping, evidenced by a 3× reduction in access to geopolitical complexity (e.g., the contested ``coup theory'') for marginalized learners; (3)~Agency Theft, a lexical shift where models like LLaMA produce a 5× higher victimization-to-politics vocabulary ratio for Roma students compared to elite peers; and (4)~Elite Hermeneutics, where AI tutors disproportionately withhold epistemic confidence and justification scores from low-resource demographic profiles. We argue that current safety alignment acts as a paternalistic filter, transforming conversational AI into agents of narrative segregation -- a manifestation of hermeneutical injustice in Fricker's~citefricker2007 sense that demands urgent pedagogical auditing.