Metraipolis › archive › dc_afdb4dd213cd

Not Safe for All: Auditing the Dialect Penalty in Text-to-Image Safety Pipelines — arXiv:2608.29589

Source: arXiv (en)
Original: https://arxiv.org/abs/2608.29589
License (as archived): CC0-1.0
Archived: 2026-10-17T07:44:21.325Z
SHA-256 of the archived text: dc3a32ea9093660a410f60abe21f9b71e8714b6547f63a5b7a75124b7cbc7136
Stable URL of this page: https://metraipolis.com/documento/dc_afdb4dd213cd

How to cite

Not Safe for All: Auditing the Dialect Penalty in Text-to-Image Safety Pipelines — arXiv:2608.29589 — arXiv (en), https://arxiv.org/abs/2608.29589 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_afdb4dd213cd (Metraipolis archive dc_afdb4dd213cd, sha256 dc3a32ea9093660a410f60abe21f9b71e8714b6547f63a5b7a75124b7cbc7136)

BibTeX

@misc{dcafdb4dd213cd,
  title = {Not Safe for All: Auditing the Dialect Penalty in Text-to-Image Safety Pipelines — arXiv:2608.29589},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.29589},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_afdb4dd213cd; sha256 dc3a32ea9093660a410f60abe21f9b71e8714b6547f63a5b7a75124b7cbc7136},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.29589. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Text-to-image (T2I) safety guardrails fail to generalize equitably to non-standard dialects. Evaluating 23,080 paired prompts across five English dialects, we formalize this failure as the dialect penalty, where filters trigger based on linguistic surface features rather than semantic intent. Text-level filters fail in opposing directions: NSFW-T over-flags benign dialect prompts and LatentGuard over-flags toxic ones (bias gaps up to +28.29 pp), while the OpenAI Moderation API under-detects them. A controlled typo ablation confirms this penalty originates from flagging dialectal features, not generic out-of-distribution sensitivity. The pixel-level generator is largely dialect-agnostic; the penalty enters at text processing and cascades unevenly to post-hoc guardrails. We show this bias tracks training data imbalance and is mitigable via group-balanced retraining, with an ablation attributing the gain to balanced exposure rather than to the worst-group objective of GroupDRO (group distributionally robust optimization). Current pipelines systematically fail dialect speakers, an equity failure masked by mean accuracy benchmarks. Our official code and dataset are publicly available at this https URL . Content Warning: This paper contains offensive, toxic, or disturbing text prompts and generated images.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.29589 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T07:28:26.557Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).