Not Safe for All: Auditing the Dialect Penalty in Text-to-Image Safety Pipelines
Text-to-image safety guardrails fail to generalize equitably to non-standard dialects, formalized as the dialect penalty, where filters trigger based on linguistic surface features rather than semantic intent, and is mitigable via group-balanced retraining.
Minkyu Kim, Juhwan Choi, Youngbin Kim
· 0 citations