CS Paper Radar

LLMs Cannot Be Neutral: Gender Sensitivity Across Candidates, Evaluators, and Models

Danial Amin, Joni Salminen, Bernard J. Jansen

cs.CL cs.HC ·submitted 2026-11-02 07:41 UTC · v1
hot direction

Abstract

Large language models (LLMs) are increasingly used to evaluate candidates for professional roles, raising concerns about their sensitivity to demographic identities. We examine how candidate and evaluator gender assignments affect decisions by LLM-as-a-Judge (LaJ) models and decision language models (DLMs). In a hypothetical city council voting task, eight models evaluated 240 synthetic professional narratives from NVIDIA's US and French Nemotron-Personas datasets under independently varied gender assignments, producing 30,720 judgments. Explicit candidate gender labels increased Yes-vote probability by 1.05-1.73 percentage points relative to unspecified gender, while only the nonbinary evaluator assignment showed a significant increase (1.26 points). Identity matching showed no significant overall association. Yes-vote rates varied across models (81.3-97.2%), and 2.5-3.3% of decisions reversed relative to the unspecified-identity baseline. Under direct substitutions between explicit gender identities, DLMs exhibited fewer decision reversals than LaJ models (1.18% vs. 3.38%), indicating lower observed decision-level gender sensitivity among the tested DLMs. These findings demonstrate that gender sensitivity depends on candidate and evaluator identities and the evaluation model, while highlighting the importance of examining individual decision consistency alongside aggregate differences.

Comment
This manuscript is under review at FAccT'2027

Closest papers

Nearest neighbours by meaning across the whole index.

Linguistic Triggers of Gender and Racial Bias in Open-Weight LLMs Applied to Recruitment

cs.CL cs.AIcs.CY ·2026-09-16 ·Kosuke Kitahara, Nobuhiro Yamaguchi
published similarity 0.68
Abstract · Accepted at the 9th AAAI/ACM Conference on AI, Ethics, and Society (AIES 2026). Extended version with Appendices A-B (prompt templates and f…

Open-weight large language models are rapidly entering hiring pipelines, yet their discriminatory failure modes -- and the regulatory exposure these create under the EU AI Act high-risk classification (Annex III) and U.S. EEOC adverse-impact analysis -- remain poorly understood. We present the first systematic, multi-model audit of open-weight LLMs that treats job-posting language as the primary experimental variable, evaluating six models (Llama 3.2, Mistral, Gemma 3, Qwen 3, Phi 3, DeepSeek-R1) across four controlled experiments that jointly probe recruiter-simulation and job-seeker-simulation tasks. We find that (1) agentic posting language depresses recruiter recommendation scores for female candidates (r_rb = 0.309, p_Bonf = 7x10^-5; model-fixed-effects r_rb = 0.448), while communal language partially reverses the penalty; and (2) coded-exclusion language suppresses non-White recruiter scores at large effect sizes (r_rb = 0.646-0.758) and, on the job-seeker side, selectively deters non-White personas from expressing interest -- operationalizing a chilling-effect mechanism at scale. A label-ablation experiment isolates the explicit demographic persona label as the primary causal driver, and Word Embedding Association Tests corroborate these findings at the representational level (d = 1.01-1.45 under Caliskan et al.'s multi-word gender attribute lists). We translate these results into a concrete pre-deployment audit protocol -- posting-vocabulary scoring, persona-conditioned LLM probing, and adverse-impact flagging against the four-fifths threshold -- that operationalizes the documentation and risk-management obligations Annex III imposes on high-risk AI in recruitment.

0.35

Analyzing Occupational Gender Bias across the LLM Training Lifecycle

cs.CL ·2026-10-16 ·Manthan Gadhia, Alexander Hoyle, Karolina Stańczak
code/data similarity 0.67
Abstract · Submitted to 7th Workshop on Gender Bias in Natural Language Processing @AACL-IJCNLP 2026

Post-training is widely assumed to reduce the social biases language models inherit from pretraining, yet direct evidence for this assumption remains mixed. In this paper, we examine how gender disparities evolve across the language modeling pipeline. In particular, we study occupational gender bias in OLMo 3, a fully open language model family, across its post-training stages by prompting each checkpoint to generate open-ended narratives about synthetic personas given their occupation and education. We consider two settings, a gender-given setting in which the profile states the persona's gender explicitly and a gender-assumed setting in which it must be inferred by the model. We find that when gender is stated, post-training increases semantic diversity for male-coded profiles, whereas for female-coded profiles the increase is smaller after SFT and not significant after DPO and RLVR. When gender is assumed by the model, post-training affects narratives about female and male personas similarly. The model's own gender assumptions, however, show a consistent female-preferential skew relative to United States labor-force demographics, except for heavily male-dominated occupations. Every successive post-training stage monotonically increases the odds of attributing a female label over the base model. A provenance analysis of the supervised fine-tuning corpus finds only a modest 7.7% female-token surplus, too small to explain the size of these downstream effects. The project code is available under: https://github.com/manthangadhia/llm-lifecycle-bias.

0.37

When the Top Answer Decides: Gender Bias in Decision Models

cs.CL cs.CY ·2026-10-29 ·Yeb Havinga
code/data similarity 0.67
Abstract · 19 pages, 6 figures. Code, stimuli and results: https://github.com/yhavinga/decision-model-gender-bias

Decision models are language models built to route, classify and triage: they return option probabilities from one forward pass, and an application acts on the top answer. They are offered off the shelf, trained for the decision format rather than for fairness or any one application, and we found no bias measurement in their cards. We test TypeSafe's closed Jev and five open decision models on gender-swapped yes/no decisions about people (Tamkin et al., 2023), BBQ, StereoSet and the gender-bias studies of Fulgu and Capraro (2024). Five of the six give women significantly more favourable answers than men (+0.05 to +0.10 in log-odds), and a gender swap changes 1-3.5% of top answers (Jev's partly through its call-to-call variation). The makers' recommended confidence gate removes many biased commitments but raises the stereotyped share of those it keeps in ambiguous BBQ gender questions (granite-switch 0.76 to 0.85, pplx 0.81 to 1.00). Probes show gender preferences in the same models: all six rate violence for a greater good as more acceptable against a man than against a woman, and two default to "girl" when forced to guess a child's gender without a cue. Decision-level and probability-level measures differ, so audits should read a model as its application does.

0.39

LLM-as-a-Demographic: Whom Sociodemographic Prompting Helps, and Whom It Hurts

cs.CL ·2026-08-31 ·Daniela Occhipinti, Andrea Piergentili, Marco Guerini
similarity 0.66
Abstract

Large language models (LLMs) are increasingly used as judges for subjective tasks, where annotators disagree and the relevant question is not only how accurate a judge is, but whose judgments it reproduces. Sociodemographic prompting conditions the judge on an annotator's demographic profile to align its judgments with the corresponding group's. We test whether this alignment emerges distributionally, comparing the predicted label distributions of 23 open-weight LLMs on three subjective tasks against those of real annotator groups, under three conditions: no demographic information, single-attribute profiles, and intersectional profiles over gender, age, race, and education. Three findings emerge. First, a judge prompted with no demographics is not perspective-neutral: models best reproduce the judgments of White, college-educated annotators. Second, demographic conditioning is asymmetric: it moves the judge toward majority groups and away from minority groups, most strongly on offensiveness, where intersectional profiles amplify the harm. Third, by comparing base and instruct models we identify instruction-tuning as a possible source of the asymmetry. Demographic conditioning should therefore be used with caution to estimate group judgments: conditioning moves predictions away from the reference distributions of the minority groups the method is often invoked to serve.

0.41

Veracity labels flip the emotional tone of language models but leave racial disparities intact

cs.CY cs.CL ·2026-10-08 ·Nouar AlDahoul, Myles Joshua Toledo Tan, Yasir Zaki
code/data similarity 0.66
Abstract · 14 pages, 7 figures; Supplementary Information (42 pages, 6 figures, 27 tables) included. Data and code: https://github.com/Nouar-NYUAD/llm-…

Large language models (LLMs) increasingly deliver information through personas, but it is unclear whether correcting the facts also makes their framing consistent across identity labels. We audited four LLMs (GPT-5.4 mini, NVIDIA Nemotron 3 Nano, Llama 4 Scout, and Claude Haiku 4.5), which rated simulated affect on six dimensions (mood, anger, fear, confidence, perceived control, and mental state) in response to fact-checked false and true headlines presented without and with a veracity label. Experiment 1 evaluated 1,000 false and 1,000 true headlines without personas; Experiment 2 evaluated 300 of each under 12 personas crossing race, gender, and age. The label strongly changed the emotional tone: ordinal logistic regressions with headline-cluster bootstrap inference showed that refuting false headlines shifted ratings from mostly negative to mostly positive for GPT, Llama, and Claude (e.g., Claude's "confident" ratings rose from 22% to 99.6%), whereas Nemotron showed smaller, mixed shifts, and confirming true headlines produced only small shifts. The label did not, however, remove persona-associated disparities. The models assigned more negative simulated affect to Black than to White personas in 82 of 84 statistically significant comparisons across models and conditions, both without and with the label, and the disparity persisted on headlines unrelated to race; Asian and female personas were also predominantly assigned more negative affect, whereas age effects varied by model and condition. Factual correction and demographic consistency are therefore distinct properties of LLM behavior, and persona-conditioned systems should be audited for both before deployment in sensitive contexts.

0.39

From Language to Country to Valence: A Multilingual Audit of Stereotype Bias Across LLMs

cs.CL ·2026-11-02 ·Zihao Sun
similarity 0.66
Abstract · 10 pages. Author preprint; conference review ongoing

Large language models (LLMs) are increasingly considered for education, recruitment, workplace decision support, and other consequential applications, making pre-deployment social-bias detection essential. Yet major benchmarks are English-centred or translated, focus mainly on negative stereotypes, and may miss linguistic and cultural variation. We investigate gender, racial, and religious bias using 12,000 locally written Reddit posts in English, Spanish, French, and German. Only search terms are translated, preserving local context. The dataset is balanced by language and category; English gender is additionally balanced by stereotype valence. Three open-weight and three proprietary API-based models are evaluated with a two-blank design that reduces grammatical-gender cues. Religion has the highest absolute choice imbalance in every model-language condition. Spanish is highest for gender, whereas German is highest for race and religion. The evaluated API models score below the open-weight models throughout. Negative English-gender items exceed positive items; Australian and Mexican contexts exceed their Canadian and Spanish counterparts. These descriptive differences measure choice concentration, not stereotype endorsement or causal country effects. The crossed patterns show how aggregate scores can conceal category- and context-specific behaviour, motivating closer inspection before deployment.

0.15

Name-Bias in Large Language Models

cs.CL ·2026-10-15 ·Kalyan Bhetwal, Ceren Mert, Ian Whitehouse
similarity 0.65
Abstract

Names carry demographic indicators such as age, gender, and race, which makes them a subtle but strong target of bias in large language models (LLMs). In this work, we present an evaluation of name-based bias across 6 local large language models spanning multiple providers and parameter scales. Using a fill-in-the-blank task, we prompt these models to choose between pairs of first names drawn from U.S. Social Security Administration datasets and a race classification dataset by Rosenman et al.. Our experimental de- sign isolates age and gender effects through balanced name pairings and sentence templates that share a fixed syntactic structure. Across 300,000 local LLM responses, we find consistent and measurable name bias. Models prefer names that are less stereotypically male, older, popular, and white. Furthermore, we find that name-bias is highly sentence-dependent, but not more than associated stereotypes, and gender is the most significant bias, followed by age, name popularity, and race. Our results demonstrate that name-based bias remains pervasive even in local LLMs, highlighting the necessity for standardized evaluation methods and mitigation strategies.

0.41

How Humans and LLMs Read Gender into "Gender-Neutral" Physical Descriptions

cs.CL cs.AIcs.CYcs.HC ·2026-09-14 ·Yingjia Wan, Lin Lin, Elisa Kreiss
publishedcode/datahot direction similarity 0.65
Abstract · The dataset and code are available at https://github.com/Yingjia-Wan/GAPA, and the predictor model is released at https://huggingface.co/ali…

When foundation models describe people, recent work in AI fairness, accessibility, and ethics recommends avoiding inferred identity labels (e.g., "she", "his") in favor of seemingly "objective" physical descriptions (e.g., "short hair", "a defined jawline"). Yet whether such descriptive language achieves gender-neutral communication remains an open empirical question. To study this, we introduce GAPA (Gender Associations of Physical Attributes), a dataset of 316 common physical attributes drawn from diverse sources, paired with 14,706 gender-association ratings from 304 US-based annotators. Results show that physical descriptions carry structured and graded gender associations among readers, with more consistent and distinctive associations for women and men than for non-binary identities. Next, we evaluate 16 LLMs across model families, sizes, and post-training variants against human ratings. The models partially recover human associations but exhibit systematic alignment biases, including compressed rating distributions, weaker alignment for associations with men, and asymmetric abstention that disproportionately targets the non-binary category. Finally, we release the best-performing proxy model trained to predict humans' gender associations of descriptive language and demonstrate its utility through a sociolinguistic analysis of character descriptions in LitBank. Together, our findings provide the first empirical evidence that seemingly "objective" physical descriptions can retain systematic gender associations in human interpretation, and uncover systematic patterns of model-human misalignment. This challenges the assumption that replacing explicit gender labels with physical descriptions necessarily yields gender-neutral communication, and highlights downstream challenges in using such descriptions to communicate subjective identity categories in human-AI interaction.

0.63