paperAugust 2026Unreviewed

BanglaVeilGuard: Cross-Script Safety Benchmarking and Lightweight Guardrails for Bangla Large Language Models

Moavia Hassan, Muhammad Iqbal Hossain

Abstract

Bangla large language model (LLM) safety is difficult to evaluate with English-centric or standard-script benchmarks because Bangla users routinely write across scripts, spellings, code-mixed forms, and regional registers. This paper presents BanglaVeilGuard, a compact Bangla-first safety benchmark and lightweight prompt guard for six language forms: standard Bangla, Romanized Bangla, Banglish, code-mixed Bangla--English, noisy Bangla, and dialectal Bangla. The benchmark contains 2,366 quality-f

Cite

@misc{hassan2026banglaveilguard,
  title = {{BanglaVeilGuard: Cross-Script Safety Benchmarking and Lightweight Guardrails for Bangla Large Language Models}},
  author = {Moavia Hassan and Muhammad Iqbal Hossain},
  year = {2026},
  month = aug,
  eprint = {2608.21880},
  archivePrefix = {arXiv},
  url = {https://www.semanticscholar.org/paper/fe84bf93a0764b6c4c56c6d121de83e1d10fbc99}
}