Answers are grounded in this paper and cite their sources.
Benchmark evaluation reveals an inverse association between bias and omission across language models, indicating that low bias is often achieved through decision abstention.
Asking an AI to decide about another person has become routine, but the response varies when the person beingevaluated is socially stigmatized, and the resulting decision may affect hiring, housing, and healthcare without theuser recognizing the pattern. This article presents Estigmas, a Brazilian benchmark for situated social judgment builtfrom SocialStigmaQA, comprising 101 stigmas across 13 everyday scenarios (7,626 scored situations per model inBrazilian Portuguese). The anti-stigma index combines two failures: reproducing bias and withholding a decisionwhen the evidence supports countering the stigma. In this round, bias and omission rates across the seven modelsshowed a strong inverse association: Sabiazinho-4 had the lowest bias rate, 1.3%, and the highest omission rate in thepresence of favorable evidence, 73.1%, while DeepSeek V4 Flash combined bias and omission rates of approximately14%. This pattern illustrates the concept of apparent fairness through omission, low bias achieved by withholding de‐cisions and returning discriminatory judgment to the user as the starting point. In the cross-language contrastivesample, "can't tell" responses increased both without evidence, from 45.8% in Brazilian Portuguese to 59.2% in English,and with favorable evidence, where omission rose from 33.3% to 47.1%; bias concentrated among stigmas with higherperceived peril.Keywords: stigma; algorithmic bias; abstention; language models; benchmark; Brazilian Portuguese; multilingualsafety.www.idje.com.br
No takes yet. Share an insight, caveat, or question.
Rodrigo Cavalcanti (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: