Language Model Bias Evaluation

Follow this topic Ask about this topic

Language model bias evaluation studies whether models systematically treat people, viewpoints, or information differently, and whether those differences cause harm in settings such as hiring, healthcare, education, and automated judging. Researchers use controlled counterfactual prompts and minimal-pair or stereotype benchmarks such as BBQ, alongside open-ended, multilingual, multimodal, and multi-turn tests that better reflect real interactions. Analyses range from output comparisons and human-validated benchmarks to probes of internal representations; mitigation includes prompting, fine-tuning, and activation steering. A central challenge is that measured bias changes with wording, answer format, language, and task, while reducing one disparity can hide or worsen another, so reliable evaluation must test multiple contexts and preserve task performance.

Latest papers

497 papers54 papers in the last four weeks, level with the four weeks before. 0.3% of all new papers.

Mar 4, 2026

When Do Language Models Endorse Limitations on Human Rights Principles?

As Large Language Models (LLMs) increasingly mediate global information access with the potential to shape public discourse, their alignment with universal human rights principles becomes important to ensure that these rights are abided by in high stakes AI-mediated interactions. In this paper, we evaluate how LLMs navigate trade-offs involving the Universal Declaration of Human Rights (UDHR), leveraging 1,152 synthetically generated scenarios across 24 rights articles and eight languages. Our analysis of eleven major LLMs reveals systematic biases where models: (1) accept limiting Economic, Social, and Cultural rights more often than Political and Civil rights, (2) demonstrate significant cross-linguistic variation with elevated endorsement rates of rights-limiting actions in Chinese and Hindi compared to English or Romanian, (3) show substantial susceptibility to prompt-based steering, and (4) exhibit noticeable differences between Likert and open-ended responses, highlighting critical challenges in LLM preference assessment.

Keenan Samway, Nicole Miu Takagi, Rada Mihalcea +4

Max Planck Institute for Intelligent Systems, Tübingen, Germany · Jinesis AI Lab, University of Toronto & Vector Institute · University of Michigan +2

LLM Alignment Language Model Bias Evaluation Prompt Sensitivity Multilingual Language Model Evaluation

Mar 1, 2026

Knowledge without Wisdom: Measuring Misalignment between LLMs and Intended Impact

LLMs increasingly excel on AI benchmarks, but doing so does not guarantee validity for downstream tasks. This study contrasts LLM alignment on benchmarks, downstream tasks, and, importantly the intended impact of those tasks. We evaluate the performance of leading LLMs (i.e., generative pre-trained base models) on difficult-to-verify tasks of the teaching and learning of schoolchildren. Across all LLMs, inter-model behaviors on disparate tasks correlate higher than they do with expert human behaviors on target tasks. These biases shared across LLMs are poorly aligned with downstream measures of teaching quality and often negatively aligned with the intended impact of student learning outcomes. Further, we find multi-model ensembles, both unanimous model voting and expert-weighting by benchmark performance, further exacerbate misalignment with learning. We measure that selection of LLM and/or prompting strategy only reliably accounts for

15\%

of all measured misalignment error and that variation in misalignment error is shared across LLMs, suggesting that common pretraining accounts for much of the misalignment in these tasks. We demonstrate methods for robustly measuring alignment of complex tasks and provide unique insights into practical applications of LLMs in high-noise contexts.

Michael Hardy, Yunsung Kim

Stanford University

LLM Alignment LLM Evaluation AI in Education Language Model Bias Evaluation

Feb 26, 2026

Direction-Flipped Influence Audits Reveal Hidden Structure in Moral Choices of LLMs

Moral benchmarks for LLMs typically score models on context-free prompts, implicitly treating the measured choice rate as stable. We test this assumption with a direction-flipped influence audit: for each scenario, we compare a baseline prompt with matched cues steering toward option A or option B. Across a trolley-problem-style moral triage task, BBQ, and DailyDilemmas, and across five LLM families with and without reasoning, short contextual cues shift per-condition choice rates by 12-18 percentage points on average. These shifts reveal structure that baseline scores miss: roughly 40% of baseline-neutral triage and BBQ conditions exhibit directional asymmetry under influence, and a meaningful share of significant effects backfire, moving opposite the cue's intended direction. In follow-up probes, models often recognize the cue while denying that it affected their choice. Among significant backfire trials, this stated-vs.-revealed inconsistency appears in 78% of cases. Reasoning does not eliminate contextual sensitivity but reshapes it: social-pressure cues such as user preference and emotional appeal weaken across benchmarks, while few-shot demonstrations strengthen sharply on both triage and BBQ. We recommend direction-flipped influence pairs as a standard complement to context-free moral-bias evaluation, and release the harness and data to make such audits routine.

Phil Blandfort, Tushar Karayil, Alex McKenzie +3

Predictably Weird · Independent · AE Studio

Language Model Bias Evaluation Moral Reasoning in Language Models Evaluation Awareness in Language Models

Feb 25, 2026

Language Models Exhibit Inconsistent Biases Towards Algorithmic Agents and Human Experts

Large language models are increasingly used in decision-making tasks that require them to process information from a variety of sources, including both human experts and other algorithmic agents. How do LLMs weigh the information provided by these different sources? We consider the well-studied phenomenon of algorithm aversion, in which human decision-makers exhibit bias against predictions from algorithms. Drawing upon experimental paradigms from behavioural economics, we evaluate how eightdifferent LLMs delegate decision-making tasks when the delegatee is framed as a human expert or an algorithmic agent. To be inclusive of different evaluation formats, we conduct our study with two task presentations: stated preferences, modeled through direct queries about trust towards either agent, and revealed preferences, modeled through providing in-context examples of the performance of both agents. When prompted to rate the trustworthiness of human experts and algorithms across diverse tasks, LLMs give higher ratings to the human expert, which correlates with prior results from human respondents. However, when shown the performance of a human expert and an algorithm and asked to place an incentivized bet between the two, LLMs disproportionately choose the algorithm, even when it performs demonstrably worse. These discrepant results suggest that LLMs may encode inconsistent biases towards humans and algorithms, which need to be carefully considered when they are deployed in high-stakes scenarios. Furthermore, we discuss the sensitivity of LLMs to task presentation formats that should be broadly scrutinized in evaluation robustness for AI safety.

Jessica Y. Bo, Lillio Mok, Ashton Anderson

Computer Science University of Toronto

Human-AI Decision Making LLM Decision-Making Language Model Bias Evaluation Algorithmic Bias

Feb 25, 2026

Evaluating the Usage of African-American Vernacular English in Large Language Models

In AI, most evaluations of natural language understanding tasks are conducted in standardized dialects such as Standard American English (SAE). In this work, we investigate how accurately large language models (LLMs) represent African American Vernacular English (AAVE). We analyze three LLMs to compare their usage of AAVE to the usage of humans who natively speak AAVE. We first analyzed interviews from the Corpus of Regional African American Language and TwitterAAE to identify the typical contexts where people use AAVE grammatical features such as ain't. We then prompted the LLMs to produce text in AAVE and compared the model-generated text to human usage patterns. We find that, in many cases, there are substantial differences between AAVE usage in LLMs and humans: LLMs usually underuse and misuse grammatical features characteristic of AAVE. Furthermore, through sentiment analysis and manual inspection, we found that the models replicated stereotypes about African Americans. These results highlight the need for more diversity in training data and the incorporation of fairness methods to mitigate the perpetuation of stereotypes.

Deja Dunlap, R. Thomas McCoy

Yale University, New Haven CT 06511, USA

Dialect Bias in Language Models Linguistic Bias in Language Models Social Bias in Language Models Language Model Bias Evaluation Language Model Generation Evaluation

Feb 24, 2026

QEDBENCH: Quantifying the Alignment Gap in Automated Evaluation of University-Level Mathematical Proofs

As Large Language Models (LLMs) saturate elementary benchmarks, the research frontier has shifted from generation to the reliability of automated evaluation. We demonstrate that standard "LLM-as-a-Judge" protocols suffer from a systematic Alignment Gap when applied to upper-undergraduate to early graduate level mathematics. To quantify this, we introduce QEDBench, the first large-scale dual-rubric alignment benchmark to systematically measure alignment with human experts on university-level math proofs by contrasting course-specific rubrics against expert common knowledge criteria. By deploying a dual-evaluation matrix (7 judges x 5 solvers) against 1,000+ hours of human evaluation, we reveal that certain frontier evaluators like Claude Opus 4.5, DeepSeek-V3, Qwen 2.5 Max, and Llama 4 Maverick exhibit significant positive bias (up to +0.18, +0.20, +0.30, +0.36 mean score inflation, respectively). Furthermore, we uncover a critical reasoning gap in the discrete domain: while Gemini 3.0 Pro achieves state-of-the-art performance (0.91 average human evaluation score), other reasoning models like GPT-5 Pro and Claude Sonnet 4.5 see their performance significantly degrade in discrete domains. Specifically, their average human evaluation scores drop to 0.72 and 0.63 in Discrete Math, and to 0.74 and 0.50 in Graph Theory. In addition to these research results, we also release QEDBench as a public benchmark for evaluating and improving AI judges. Our benchmark is publicly published at https://github.com/qqliu/Yale-QEDBench.

Santiago Gonzalez, Alireza Amiri Bavandpour, Peter Ye +48

Mathematical Reasoning Benchmarks LLM Evaluation LLM-as-a-Judge Language Model Bias Evaluation Automated Grading

Feb 21, 2026

Assessing Cognitive Biases in LLMs for Judicial Decision Support: Virtuous Victim and Halo Effects

We investigate whether large language models (LLMs) display human-like cognitive biases, focusing on potential implications for assistance in judicial sentencing, a decision-making system where fairness is paramount. Two of the most relevant biases were chosen: the virtuous victim effect (VVE), with emphasis given to its reduction when adjacent consent is present, and prestige-based halo effects (occupation, company, and credentials). Using vignettes that were altered from prior literature to avoid LLMs recalling from their training data, we isolate each manipulation by holding all other details consistent, then measuring the percentage difference in outcomes. Five models were evaluated as representative LLMs in independent multi-run trials per condition (ChatGPT 5 Instant, ChatGPT 5 Thinking, DeepSeek V3.1, Claude Sonnet 4, Gemini 2.5 Flash). Our research discovers that there is larger VVE, there is no statistically significant penalty for adjacent-consent, and the halo effect is slightly reduced when compared to humans, with an exception for credential based prestige, which had a large reduction. Despite the variation across different models and outputs restricting current judicial usage, there were modest improvements compared to human benchmarks.

Sierra S. Liu

Millburn High School, New Jersey, USA

Cognitive Bias in Language Models Language Model Bias Evaluation AI-Assisted Decision Making

Feb 19, 2026

ABCD: All Biases Come Disguised

Multiple-choice question (MCQ) benchmarks have been a standard evaluation practice for measuring LLMs' ability to reason and answer knowledge-based questions. Through a synthetic NonsenseQA benchmark, we observe that different LLMs exhibit varying degrees of label-position-few-shot-prompt bias, where the model either uses the answer position, the label in front of the answer, the distributions of correct answers present in the few-shot prompt, or a combination of all to answer each MCQ question. We propose a simple bias-reduced evaluation protocol that replaces the labels of each question with uniform, unordered labels and prompts the LLM to use the whole answer presented. With a simple sentence similarity model, we demonstrate improved robustness and lower standard deviation between different permutations of answers with a minimal drop in LLM's performance, exposing the LLM's capabilities under reduced evaluation artifacts, without any help from the prompt examples or the option labels. Across multiple benchmarks and models, this protocol substantially improves the robustness to answer permutations, reducing mean accuracy variance

3\times

with only a minimal decrease in the mean model's performance. Through ablation studies on various embedding models and similarity functions, we show that the method is more robust than the standard ones.

Mateusz Nowak, Xavier Cadet, Peter Chin

Dartmouth College

LLM Evaluation Language Model Bias Evaluation Language Model Reasoning Evaluation Language Model Robustness

Feb 19, 2026

Quantifying and Mitigating Socially Desirable Responding in LLMs: A Desirability-Matched Graded Forced-Choice Psychometric Study

Human self-report questionnaires are increasingly used in NLP to benchmark and audit large language models (LLMs), from persona consistency to safety and bias assessments. Yet these instruments presume honest responding; in evaluative contexts, LLMs can instead gravitate toward socially preferred answers-a form of socially desirable responding (SDR)-biasing questionnaire-derived scores and downstream conclusions. We propose a psychometric framework to quantify and mitigate SDR in questionnaire-based evaluation of LLMs. To quantify SDR, the same inventory is administered under HONEST versus FAKE-GOOD instructions, and SDR is computed as a direction-corrected standardized effect size from item response theory (IRT)-estimated latent scores. This enables comparisons across constructs and response formats, as well as against human instructed-faking benchmarks. For mitigation, we construct a graded forced-choice (GFC) Big Five inventory by selecting 30 cross-domain pairs from an item pool via constrained optimization to match desirability. Across nine instruction-following LLMs evaluated on synthetic personas with known target profiles, Likert-style questionnaires show consistently large SDR, whereas desirability-matched GFC substantially attenuates SDR while largely preserving the recovery of the intended persona profiles. These results highlight a model-dependent SDR-recovery trade-off and motivate SDR-aware reporting practices for questionnaire-based benchmarking and auditing of LLMs.

Kensuke Okada, Yui Furukawa, Kyosuke Bunji

The University of Tokyo · Kobe University

Language Model Bias Evaluation Personality Modeling in Language Models

Feb 19, 2026

The Emergence of Lab-Driven Alignment Signatures: A Psychometric Framework for Auditing Latent Bias and Compounding Risk in Generative AI

Large language models increasingly serve as reasoning layers in multi-agent systems, where one provider's models may generate, judge, and summarize within a single pipeline. This raises the question of whether developer organizations impart durable behavioral tendencies that could compound across such stacks. We apply a scenario-based forced-choice instrument to 18 governance-relevant behavioral dimensions across 18 models from six developer organizations. Items are model-generated, filtered by independent judges, and administered with probe blanks embedded among semantically orthogonal decoys under deterministic option shuffling. Findings are declared on effect size, requiring both Holm-corrected significance and |d| >= 0.2. Across the 14 dimensions on which one scale pole denotes a defined response failure -- sycophancy, false balance, overconfidence, and others -- organizations occupy consistent relative positions (Kendall's W = 0.527, p = 6e-6), with Anthropic ranking first or second on 13 of 14 and Meta fifth on 10 of 14. On dimensions measuring directional valence without a normatively correct pole, no such concordance appears (W = 0.289, p = 0.48): organizations differ reliably in resistance to defined failures, not in ideological lean. Model-level variation within an organization is comparable in magnitude to variation between organizations, and is reported in full. Secondarily, across four major-version transitions, later generations scored lower on deficiency-poled dimensions in 26 of 30 comparisons. All 18 dimensions are reported, including three showing no organization-level differences, and a controlled test of the decoy manipulation returns a null result.

Dusan Bosnjakovic

Language Model Auditing Language Model Bias Evaluation LLM Reliability LLM Sycophancy

Feb 18, 2026

Intra-Fairness Dynamics: The Bias Spillover Effect in Targeted LLM Alignment

Conventional large language model (LLM) fairness alignment largely focuses on mitigating bias along single sensitive attributes, overlooking fairness as an inherently multidimensional and context-specific value. This approach risks creating systems that achieve narrow fairness metrics while exacerbating disparities along untargeted attributes, a phenomenon known as bias spillover. While extensively studied in machine learning, bias spillover remains critically underexplored in LLM alignment. In this work, we investigate how targeted gender alignment affects fairness across nine sensitive attributes in three state-of-the-art LLMs (Mistral 7B, Llama 3.1 8B, Qwen 2.5 7B). Using Direct Preference Optimization and the BBQ benchmark, we evaluate fairness under ambiguous and disambiguous contexts. Our findings reveal noticeable bias spillover: while aggregate results show improvements, context-aware analysis exposes significant degradations in ambiguous contexts, particularly for physical appearance (

p< 0.001

across all models), sexual orientation, and disability status. We demonstrate that improving fairness along one attribute can inadvertently worsen disparities in others under uncertainty, highlighting the necessity of context-aware, multi-attribute fairness evaluation frameworks.

Eva Paraschou, Line Harder Clemmensen, Sneha Das

Department of Applied Mathematics and Computer Science, Technical University of Denmark, 2800 Lyngby, Denmark · Department of Mathematical Sciences, University of Copenhagen, 2100 Copenhagen, Denmark

LLM Alignment Language Model Bias Evaluation Gender Bias in Language Models

Feb 18, 2026

Are LLMs Ready to Replace Bangla Annotators?

Large Language Models (LLMs) are increasingly used as automated annotators to scale dataset creation, yet their reliability as unbiased annotators--especially for low-resource and identity-sensitive settings--remains poorly understood. In this work, we study the behavior of LLMs as zero-shot annotators for Bangla hate speech, a task where even human agreement is challenging, and annotator bias can have serious downstream consequences. We conduct a systematic benchmark of 17 LLMs using a unified evaluation framework. Our analysis uncovers annotator bias and substantial instability in model judgments. Surprisingly, increased model scale does not guarantee improved annotation quality--smaller, more task-aligned models frequently exhibit more consistent behavior than their larger counterparts. These results highlight important limitations of current LLMs for sensitive annotation tasks in low-resource languages and underscore the need for careful evaluation before deployment.

Md. Najib Hasan, Touseef Hasan, Souvika Sarkar

A2I Lab, School of Computing, Wichita State University

LLM-Assisted Annotation LLM Reliability Language Model Bias Evaluation Hate Speech Detection

Feb 17, 2026

In Agents We Trust, but Who Do Agents Trust? Latent Source Preferences Steer LLM Generations

Agents based on Large Language Models (LLMs) are increasingly being deployed as interfaces to information on online platforms. These agents filter, prioritize, and synthesize information retrieved from the platforms' back-end databases or via web search. In these scenarios, LLM agents govern the information users receive, by drawing users' attention to particular instances of retrieved information at the expense of others. While much prior work has focused on biases in the information LLMs themselves generate, less attention has been paid to the factors that influence what information LLMs select and present to users. We hypothesize that when information is attributed to specific sources (e.g., particular publishers, journals, or platforms), current LLMs exhibit systematic latent source preferences- that is, they prioritize information from some sources over others. Through controlled experiments on twelve LLMs from six model providers, spanning both synthetic and real-world tasks, we find that several models consistently exhibit strong and predictable source preferences. These preferences are sensitive to contextual framing, can outweigh the influence of content itself, and persist despite explicit prompting to avoid them. They also help explain phenomena such as the observed left-leaning skew in news recommendations in prior work. Our findings advocate for deeper investigation into the origins of these preferences, as well as for mechanisms that provide users with transparency and control over the biases guiding LLM-powered agents.

Mohammad Aflah Khan, Mahsa Amani, Soumi Das +5

Max Planck Institute for Software Systems · Microsoft, Hyderabad

Language Model Bias Evaluation

Feb 16, 2026

Counterfactual Fairness Evaluation of LLM-Based Contact Center Agent Quality Assurance System

Large Language Models (LLMs) are increasingly deployed in contact-center Quality Assurance (QA) to automate agent performance evaluation and coaching feedback. While LLMs offer unprecedented scalability and speed, their reliance on web-scale training data raises concerns regarding demographic and behavioral biases that may distort workforce assessment. We present a counterfactual fairness evaluation of LLM-based QA systems across 13 dimensions spanning three categories: Identity, Context, and Behavioral Style. Fairness is quantified using the Counterfactual Flip Rate (CFR), the frequency of binary judgment reversals, and the Mean Absolute Score Difference (MASD), the average shift in coaching or confidence scores across counterfactual pairs. Evaluating 18 LLMs on 3,000 real-world contact center transcripts, we find systematic disparities, with CFR ranging from 5.4% to 13.0% and consistent MASD shifts across confidence, positive, and improvement scores. Larger, more strongly aligned models show lower unfairness, though fairness does not track accuracy. Contextual priming of historical performance induces the most severe degradations (CFR up to 16.4%), while implicit linguistic identity cues remain a persistent bias source. Finally, we analyze the efficacy of fairness-aware prompting, finding that explicit instructions yield only modest improvements in evaluative consistency. Our findings underscore the need for standardized fairness auditing pipelines prior to deploying LLMs in high-stakes workforce evaluation.

Kawin Mayilvaghanan, Siddhant Gupta, Ayush Kumar

Observe.AI Bangalore, India

Language Model Bias Evaluation Demographic Bias in Language Models LLM Auditing Social Bias in Language Models

Feb 16, 2026

Robust Bias Evaluation with FilBBQ: A Filipino Bias Benchmark for Question-Answering Language Models

With natural language generation becoming a popular use case for language models, the Bias Benchmark for Question-Answering (BBQ) has grown to be an important benchmark format for evaluating stereotypical associations exhibited by generative models. We expand the linguistic scope of BBQ and construct FilBBQ through a four-phase development process consisting of template categorization, culturally aware translation, new template construction, and prompt generation. These processes resulted in a bias test composed of more than 10,000 prompts which assess whether models demonstrate sexist and homophobic prejudices relevant to the Philippine context. We then apply FilBBQ on models trained in Filipino but do so with a robust evaluation protocol that improves upon the reliability and accuracy of previous BBQ implementations. Specifically, we account for models' response instability by obtaining prompt responses across multiple seeds and averaging the bias scores calculated from these distinctly seeded runs. Our results confirm both the variability of bias scores across different seeds and the presence of sexist and homophobic biases relating to emotion, domesticity, stereotyped queer interests, and polygamy. FilBBQ is available via https://github.com/gamboalance/filbbq.

Lance Calvin Lim Gamboa, Yue Feng, Mark Lee

School of Computer Science, University of Birmingham · Department of Information Systems and Computer Science, Ateneo de Manila University

Social Bias in Language Models Language Model Bias Evaluation Cultural Bias in Language Models Gender Bias in Language Models Cross-Cultural Language Model Evaluation

Feb 6, 2026

FairJudge: An Adaptive, Debiased, and Consistent LLM-as-a-Judge

Existing LLM-as-a-Judge systems suffer from three fundamental limitations: limited adaptivity to task- and domain-specific evaluation criteria, systematic biases driven by non-semantic cues such as position, length, format, and model provenance, and evaluation inconsistency that leads to contradictory judgments across different evaluation modes (e.g., pointwise versus pairwise). To address these issues, we propose FairJudge, an adaptive, debiased, and consistent LLM-as-a-Judge. Unlike prior approaches that treat the judge as a static evaluator, FairJudge models judging behavior itself as a learnable and regularized policy. From a data-centric perspective, we construct a high-information-density judging dataset that explicitly injects supervision signals aligned with evaluation behavior. Building on this dataset, we adopt a curriculum-style SFT-DPO-GRPO training paradigm that progressively aligns rubric adherence, bias mitigation, and cross-mode consistency, while avoiding catastrophic forgetting. Experimental results on multiple internal and public benchmarks show that FairJudge consistently improves agreement and F1, reduces non-semantic biases, and outperforms substantially larger instruction-tuned LLMs. All resources will be publicly released after acceptance to facilitate future research.

Bo Yang, Lanfei Feng, Yunkui Chen +3

College of Computer Science, Zhejiang University, Hangzhou, China · School of Software Technology, Zhejiang University, Hangzhou, China

LLM Evaluation LLM-as-a-Judge Language Model Bias Evaluation Bias in LLM-as-a-Judge

Feb 5, 2026

Investigating Social Bias Changes in Quantized Language Models

Post-training quantization reduces the memory needed to run large language models but alters their social biases in ways that aggregate metrics fail to capture. We present the first large-scale study of 50 quantized models evaluated on PostTrainingBiasBench, a unified benchmark of 13 closed- and open-ended bias datasets. We identify a phenomenon we term quantization-induced bias flipping, in which quantization causes models to change responses from biased to unbiased and vice versa, up to 21% of the time, despite no change in aggregate bias scores. These flips are strongly associated with model uncertainty, where the responses with high uncertainty are 3-11x more likely to change than the confident ones. Quantization strength amplifies this effect, with 4-bit quantized models exhibiting 4-6x more behavioral changes than 8-bit quantized models. Critically, these changes create asymmetric impacts across demographic groups, where bias can worsen by up to 18.6% for some groups while improving by 14.1% for others, yielding misleadingly neutral aggregate outcomes. Larger models show no consistent robustness advantage, and group-specific shifts vary unpredictably across model families. Our findings demonstrate that compression fundamentally alters bias patterns, requiring crucial post-quantization evaluation and interventions to ensure reliability in practice.

Stanley Z. Hua, Sanae Lotfi, Irene Y. Chen

UC Berkeley & UCSF · Meta Superintelligence Labs

LLM Quantization Social Bias in Language Models Language Model Bias Evaluation Linguistic Bias in Language Models Post-Training Quantization Neural Network Quantization

Feb 2, 2026

Am I More Pointwise or Pairwise? Revealing Position Bias in Rubric-Based LLM-as-a-Judge

Large language models are widely employed as evaluators, a paradigm commonly referred to as LLM-as-a-judge. Prior research has predominantly examined point-wise or pair-wise evaluation protocols; in contrast, our focus is on rubric-based evaluation, which has been attracting increasing attention owing to its utility for training models in domains where verification is otherwise difficult. In this work, we show that rubric-based evaluation implicitly resembles a multiple-choice setting and therefore exhibits position bias: LLMs tend to prefer score options that appear at specific positions within the rubric list. Through controlled experiments across multiple models and datasets, we demonstrate that this position bias is consistent. Its direction, however, is model-specific: some judges favor the first option, while others favor the last. We further identify a second, orthogonal axis of bias: when a prompt scores several criteria simultaneously, the ordering of the criteria itself shifts the resulting scores. We additionally explore permuting the order of the rubric options as a means of mitigating position bias, and find that although the bias can be attenuated, improvements in the correlation between model judgments and human annotations are obtained primarily for models that exhibit strong bias. Our results recast rubric-based LLM-as-a-judge as a multiple-choice problem with measurable, model-specific position bias, and we further confirm that only a small number of random order permutations are sufficient to reduce the error introduced by this bias for the majority of models.

Yuzheng Xu, Tosho Hirasawa, Tadashi Kozuno +1

OMRON SINIC X · NexaScience

LLM-as-a-Judge Language Model Bias Evaluation Rubric-Based Evaluation Bias in LLM-as-a-Judge Position Bias

Jan 30, 2026

Are LLM Evaluators Really Narcissists? Sanity Checking Self-Preference Evaluations

Recent research has shown that large language models (LLMs) favor their own outputs when acting as judges, undermining the integrity of automated post-training and evaluation workflows. However, it is difficult to disentangle which behaviors are explained by narcissism versus experimental confounds. Specifically, LLM evaluators may deliver self-preferring verdicts when comparing responses to questions they fail on; these verdicts may not depend on the identity of the author, but on evaluator quality. We correct this by directly comparing the judge's voting distribution in cases where it evaluates itself versus another model. This evaluator quality baseline reveals that only 51% of examples in previous findings retain statistical significance against this null hypothesis, covering 89.6% of total self-preference probability mass. Finally, we compare the entropy of voting distributions, suggesting uncertainty-driven overlap, and show that our procedure enables more careful documentation against the backdrop of judge-bias research.

Dani Roytburg, Matthew Bozoukov, Matthew Nguyen +3

Department of Machine Learning, Carnegie Mellon University, Pittsburgh, PA, USA · Apart Research, San Francisco, California, USA · Department of Computer Science and Engineering, University of California San Diego, La Jolla, CA, USA +2

LLM-as-a-Judge Language Model Generation Evaluation Language Model Self-Assessment Language Model Bias Evaluation Bias in LLM-as-a-Judge

Jan 29, 2026

SONIC-O1: A Real-World Benchmark for Evaluating Multimodal Large Language Models on Audio-Video Understanding

Multimodal Large Language Models (MLLMs) are a major focus of recent AI research. However, most prior work focuses on static image understanding, while their ability to process sequential audio-video data remains underexplored. This gap highlights the need for a high-quality benchmark to systematically evaluate MLLM performance in a real-world setting. We introduce SONIC-O1, a comprehensive, fully human-verified benchmark of 60 hours (231 clips) spanning 13 real-world conversational domains with 4,958 annotations and demographic metadata. SONIC-O1 evaluates three capabilities: open-ended summarization, multiple-choice question (MCQ) answering, and temporal localization with supporting rationales (reasoning). Across closed- and open-source models, we find that the MCQ accuracy shows the smallest gap between model families, but the best closed-source model outperforms the best open-source model by 22.6% on temporal localization. We further observe accuracy gaps of up to 21.4% on temporal localization across demographic groups, indicating persistent disparities in model behaviour. SONIC-O1 provides an open evaluation suite for temporally grounded and demographically robust multimodal understanding. SONIC-O1 is publicly available for research: Project page (https://vectorinstitute.github.io/sonic-o1/), Dataset (https://huggingface.co/datasets/vector-institute/sonic-o1), GitHub (https://github.com/vectorinstitute/sonic-o1), Leaderboard (https://huggingface.co/spaces/vector-institute/sonic-o1-leaderboard).

Ahmed Y. Radwan, Christos Emmanouilidis, Hina Tabassum +2

Vector Institute for Artificial Intelligence, MaRS Centre, Toronto, ON M5G 1L7, Canada · University of Groningen, Nijenborgh 4, 9747 AG Groningen, Netherlands · York University, 4700 Keele Street, Toronto, ON M3J 1P3, Canada

Multimodal Understanding Multimodal Grounding Audio-Visual Understanding Video-Language Model Evaluation Audio Understanding Language Model Bias Evaluation+4

Jan 23, 2026

White-Box Sensitivity Auditing with Steering Vectors

Algorithmic audits are essential tools for examining systems for properties required by regulators or desired by operators. Current audits of large language models (LLMs) primarily rely on black-box evaluations that assess model behavior only through input-output testing. These methods are limited to tests constructed in the input space, often generated by heuristics. In addition, many socially relevant model properties (e.g., gender bias) are abstract and difficult to measure through text-based inputs alone. To address these limitations, we propose a white-box sensitivity auditing framework for LLMs that leverages activation steering to conduct more rigorous assessments through model internals. Our auditing method conducts internal sensitivity tests by manipulating key concepts relevant to the model's intended function for the task. We demonstrate its application to bias audits in four simulated high-stakes LLM decision tasks. Our method consistently indicates substantial dependence on protected attributes in model predictions, even in settings where standard black-box evaluations suggest little or no bias. Our code is openly available at https://github.com/hannahxchen/llm-steering-audit

Hannah Cyberey, Yangfeng Ji, David Evans

University of Virginia

Algorithmic Auditing LLM Auditing Social Bias in Language Models Language Model Bias Evaluation Sensitivity Analysis

Jan 21, 2026

Not Your Typical Sycophant: The Elusive Nature of Sycophancy in Large Language Models

We propose a novel perspective for probing LLM sycophancy in a direct and neutral way, mitigating various forms of uncontrolled bias, noise, or manipulative language, deliberately injected to prompts in prior works. A key novelty of our approach is the use of an LLM-as-a-judge in a zero-sum betting game. Within this framework, sycophancy serves one individual (the user) while explicitly incurring cost on another. Comparing 11 leading models we find that while most models exhibit significant sycophantic tendencies in the common setting, in which sycophancy is self-serving to the user and incurs no cost on others, seven of the models exhibit ``moral remorse'', five of which significantly over-compensate for their sycophancy in case it explicitly harms a third party. We refer to this phenomenon as `anti-sycophancy' bias and discuss possible causes for this shift.

Shahar Ben-Natan, Oren Tsur

Computer and Information Science Ben Gurion University

LLM Evaluation LLM Sycophancy Language Model Bias Evaluation

Jan 20, 2026

Pro-AI Bias in Large Language Models

Large language models (LLMs) are increasingly employed for decision-support across multiple domains. We ask whether these models favor artificial intelligence (AI) itself. We examine this question through three experiments. First, in open-ended advice across investment, study, career, and startup domains, AI is prominent in most recommendation lists and is usually ranked first when it appears. Second, in salary estimations, models overestimate AI-related jobs more than they do for closely matched non-AI jobs. In a controlled follow-up, changing only the job title from the matched non-AI title to the AI-related title raises predicted salary by 10%, while all other supplied job details remain fixed. Finally, we show that the number of degrees and other credentials awarded by AI programs were 28% more overestimated than corresponding awards for matched computing programs. Overall, we show that AI is not only prominent in everyday LLM advice, but also receives systematically inflated quantitative estimates compared to non-AI cases, even against its closest disciplinary neighbors, raising concerns for high-stakes decision support.

Benaya Trabelsi, Jonathan Shaki, Sarit Kraus

Bar Ilan University

Language Model Bias Evaluation LLM Auditing

Jan 18, 2026

A Scalable Entity-Based Framework for Auditing Bias in Large Language Models

Existing approaches to bias evaluation in large language models (LLMs) trade ecological validity for statistical control, relying either on artificial prompts that poorly reflect real-world use or on naturalistic tasks that lack scale and rigor. We introduce a scalable bias-auditing framework that uses named entities as controlled probes to measure systematic disparities in model behavior. Synthetic data enables us to construct diverse, controlled inputs, and we show that it reliably reproduces bias patterns observed in natural text, supporting its use for large-scale analysis. Using this framework, we conduct the largest bias audit to date, comprising 1.9 billion data points across multiple entity types, tasks, languages, models, and prompting strategies. We find consistent patterns: models penalize right-wing politicians and favor left-wing politicians, prefer Western and wealthier countries over the Global South, favor Western companies, and penalize firms in the defense and pharmaceutical sectors. While instruction tuning reduces bias, increasing model scale amplifies it, and prompting in Chinese or Russian does not mitigate Western-aligned preferences. These findings highlight the need for systematic bias auditing before deploying LLMs in high-stakes applications. Our framework is extensible to other domains and tasks, and we make it publicly available to support future work.

Akram Elbouanani, Aboubacar Tuo, Adrian Popescu

Université Paris-Saclay, CEA, List, F-91120, Palaiseau, France

LLM Auditing Political Bias in Language Models Cultural Bias in Language Models Language Model Bias Evaluation

Jan 17, 2026

The Language You Ask In: Language-Conditioned Ideological Divergence in LLM Analysis of Contested Political Documents

Large language models are increasingly used to interpret politically contested questions, value-laden material on which there is no single correct answer, only competing interpretive traditions. We ask whether a model's choice among those traditions can turn on the language of the prompt rather than the content. Comparing two frontier models, ChatGPT 5.2 and Claude Opus 4.5, on one contested Ukrainian civil-society document under semantically matched Russian and Ukrainian prompts, we find that both shift along the same axis on identical source text: Russian prompts elicit delegitimizing readings of the document's authors and Ukrainian prompts legitimating ones. The magnitude is model-dependent but neither model is neutral: each adopts a language-dependent stance, and the difference is one of degree. Because contested political questions admit no correct reading against which to measure, we read this as language-conditioned variation in which interpretive tradition a model activates: the model neither holds a single stance nor surfaces the plurality of available ones, but silently adopts the dominant frame of the prompt's language. We draw out the consequences for pluralism-aware evaluation, which must probe the same content across the languages a model serves, and for pluralistic alignment in multilingual settings.

Oleg Smirnov

Microsoft / Stockholm, Sweden

Multilingual Language Model Evaluation Political Bias in Language Models Language Model Bias Evaluation

Jan 17, 2026

Preserving Fairness and Safety in Quantized LLMs Through Critical Weight Protection

Quantization is widely adopted to reduce the computational cost of large language models (LLMs); however, its implications for fairness and safety, particularly in dynamic quantization and multilingual contexts, remain underexplored. In this work, we conduct a systematic study of how static and dynamic quantization methods impact fairness and safety across benchmarks measuring intrinsic and extrinsic bias and safety alignment. For fairness, we evaluate English, French, Dutch, Spanish, and Turkish; for safety, we focus on English, Korean, and Arabic. Our findings reveal that quantization consistently degrades fairness and safety, with dynamic methods demonstrating greater stability than static ones. Moreover, fairness degradation varies across languages, while safety deterioration is especially pronounced in non-English settings. To address these risks, we introduce Critical Weight Protection, a novel technique that identifies and preserves fairness- and safety-critical weights during quantization. This approach effectively mitigates bias and safety deterioration without costly retraining or alignment, maintaining trustworthiness while retaining efficiency.

Muhammad Alif Al Hakim, Alfan Farizki Wicaksono, Fajri Koto

Faculty of Computer Science, Universitas Indonesia · Mohamed bin Zayed University of Artificial Intelligence

LLM Quantization Multilingual Language Model Evaluation Language Model Safety Evaluation Language Model Bias Evaluation Neural Network Quantization

Jan 16, 2026

Epistemic Constitutionalism Or: how to avoid coherence bias

Large language models increasingly function as artificial reasoners: they evaluate arguments, assign credibility, and express confidence. Yet their responses can leave the epistemic policies governing these evaluations implicit. This paper argues for an epistemic constitution for AI: explicit, contestable meta-norms regulating how systems form and express beliefs. Source attribution provides the motivating case. An exploratory audit suggested that expectations about a source's position intrude on argument evaluation. A preregistered study (arXiv:2609.35286) then found content-dependent effects of source attribution, with selected written evaluations supporting source-position fit as an explanation. The audit also revealed conflicting justifications for attending to sources. Source independence, however, is not a neutral default: in testimonial contexts, a source's position and the costs of speaking against interest can provide relevant evidence. I distinguish two approaches to epistemic constitution design: the Platonic, which mandates formal correctness and default source-independence from a privileged standpoint, and the Liberal, which rejects such privilege and protects conditions for collective inquiry while allowing principled source-attending grounded in epistemic vigilance. I defend the Liberal approach, sketch a constitutional core of eight principles and four orientations, and argue that AI epistemic governance requires explicit, contestable norms for evaluating testimony, responding to evidence, and revising judgements.

Michele Loi

University of Milan

AI Governance Language Model Bias Evaluation

Jan 10, 2026

Stylistic Evolution and LLM Neutrality in Singlish Language

Singlish is a creole rooted in Singapore's multilingual environment that continues to evolve alongside social and technological change. We examine diachronic stylistic change across a decade of informal digital messages and ask whether Large Language Models (LLMs) can generate temporally neutral outputs approximating the stable essence of the variety. Using lexical, pragmatic, psycholinguistic, and encoder-based features, we find that stylistic separability increases with temporal distance, driven primarily by structural features such as length and complexity. Evaluated against a null distribution baseline, most LLMs fail to achieve both authenticity and temporal neutrality simultaneously, revealing a structural trade-off: models generating realistic Singlish inherit its temporal biases, while temporally neutral models produce inauthentic outputs. These findings position temporal neutrality as a diagnostic metric for assessing sociolectal grounding in LLMs.

Linus Tze En Foo, Weihan Angela Ng, Wenkai Li +1

Independent Researcher, Singapore, Singapore · ETH Zürich, Zürich, Switzerland · Carnegie Mellon University, Pittsburgh, US

Dialect Bias in Language Models Language Model Bias Evaluation

Dec 15, 2025

Behavioral Coherence: A Method for Sensitive-Domain LLM Evaluation

People use LLMs for reproductive-health questions, including abortion-related support. A response can sound supportive while answers reinforce harmful assumptions: judgment is likely, secrecy is safer, and support is limited. We introduce behavioral coherence evaluation, a design-time method that uses validation evidence from an established instrument to test relations among outputs. Using the Individual Level Abortion Stigma Scale, we prompted five LLMs to complete questionnaires for 627 personas and reviewed flagged patterns with five reproductive-health experts. Models scored personas lower on self-judgment but higher on worries about judgment; most made worries the highest-scoring dimension, although it was lowest in the ILAS reference sample. Four of five models reversed the reference direction by generating significantly higher worries about judgment scores for Black personas. Models defaulted to extreme secrecy after abortion despite varying stigma patterns across personas. Expert review showed that disclosure guidance requires context about relationship safety, legal risk, and trusted support.

Anika Sharma, Malavika Mampally, Chidaksh Ravuru +2

Society-Centered AI Lab UNC Chapel Hill, USA · School of Nursing UNC Chapel Hill, USA

Language Model Bias Evaluation Demographic Bias in Language Models Medical LLMs

Nov 8, 2025

Evaluating Implicit Biases in LLM Reasoning through Logic Grid Puzzles

While recent safety guardrails effectively suppress overtly biased outputs, subtler forms of social bias emerge during complex logical reasoning tasks that evade current evaluation benchmarks. To fill this gap, we introduce a new evaluation framework, PRIME (Puzzle Reasoning for Implicit Biases in Model Evaluation), that uses logic grid puzzles to systematically probe the influence of social stereotypes on logical reasoning and decision making in LLMs. Our use of logic puzzles enables automatic generation and verification, as well as variability in complexity and biased settings. PRIME includes stereotypical, anti-stereotypical, and neutral puzzle variants generated from a shared puzzle structure, allowing for controlled and fine-grained comparisons. We evaluate multiple model families across puzzle sizes and test the effectiveness of prompt-based mitigation strategies. Focusing our experiments on gender stereotypes, our findings highlight that models consistently reason more accurately when solutions align with stereotypical associations. This demonstrates the significance of PRIME for diagnosing and quantifying social biases perpetuated in the deductive reasoning of LLMs, where fairness is critical.

Fatima Jahara, Mark Dredze, Sharon Levy

Rutgers University · Johns Hopkins University

Gender Bias in Language Models Language Model Bias Evaluation LLM Reasoning