Bias in LLM-as-a-Judge

Ask about this topic

Overview

This field examines how automated judges can favor certain answers for reasons unrelated to quality, including response order, writing style, verbosity, model identity, language, and demographic or cultural cues. Researchers use controlled counterfactual tests and benchmarks such as MT-Bench, RewardBench, and multilingual or domain-specific evaluation sets to measure when judgments change and whether they match human ratings. Mitigation approaches include randomizing presentation, adding rubrics and reference answers, calibrating scores with human comparisons, combining judges robustly, and training models to separate relevant evidence from distracting cues. A central challenge is that repeatable judgments and agreement among models do not guarantee validity, while interventions that reduce bias can also erase real quality differences or legitimate differences in human preferences.

Latest papers

CardsList

118 papers21 papers in the last four weeks, up 11% on the four weeks before. 0.1% of all new papers.

Jul 2, 2026

Ask the Right Comparison:Bias-Aware Bayesian Active Top-kk Ranking with LLM Judges

Large language models (LLMs) are increasingly used as cheap, scalable judges that compare candidate outputs pairwise -- to rank responses, select models, or triage papers. Yet LLM judges are both noisy and systematically biased: they favor verbose or well-formatted answers and exhibit position effects, so simply aggregating their votes recovers a ranking of presentation, not of true quality. We study the practical goal of identifying the \topk{} items under a fixed comparison budget, and make two contributions. First, we cast judging as Bayesian inference over latent quality with explicit, judge-specific bias covariates (verbosity, position), regularized by a shrinkage prior so that the data decide which biases a given judge actually exhibits. Second, we introduce a \topk-aware active acquisition rule that chooses the next comparison to maximally reduce uncertainty about \topk{} membership, rather than about the full ranking. On a controlled benchmark with known ground-truth quality, judged by sixteen real LLMs spanning open and proprietary families (Llama, Qwen, Phi-4, GPT-4o-mini/5.1/5.5, Gemini, DeepSeek, and Claude Haiku/Sonnet/Opus), naive aggregation plateaus at a wrong \topk{} on biased judges regardless of budget, while our bias-aware model recovers it; \topk-aware acquisition reaches this ceiling with far fewer comparisons than round-robin or a global-uncertainty (D-optimal) rule. Bias is real but heterogeneous and capability-dependent: cheap and mid-tier judges carry a strong verbosity bias that our model corrects (lifting recall from \sim$$0.5--0.60.6 to 0.840.84--1.01.0), whereas the frontier judges we tested show little bias and already rank accurately, so bias-aware modeling changes little there.

Jian Xu, Delu Zeng, John Paisley +1

RIKEN iTHEMS · RIKEN AIP · South China University of Technology +1

Pairwise ComparisonLLM-as-a-JudgeLearning to RankLLM RerankingPairwise RankingLanguage Model Bias Evaluation+2
AskPDF
Jun 29, 2026

RoPoLL: Robust Panel of LLM Judges

The LLM Jury, a Panel of LLM Evaluators (PoLL) reporting consensus scores, has become a practical alternative to single-judge LLM evaluation, yet its statistical behavior remains poorly understood. We formalize the LLM Jury under the Huber contamination model and show that PoLL incurs unbounded bias under any positive contamination, regardless of jury size, whenever a single judge fails in a biased, LLM-typical way (mode collapse, sycophancy, safety refusal). Framing jury consensus as classical robust mean estimation, we propose RoPoLL (Robust Panel of LLM-as-Judge), which preserves the PoLL panel but replaces the aggregation function with a robust mean estimator, instantiated with the geometric median (GM): tuning-free, with the optimal finite-sample breakdown point 1/2. A finite-sample error bound and a matching information-theoretic minimax lower bound agree on the parametric rate sigma*sqrt(d/N) and differ on the breakdown floor by a factor of sqrt(d), a statistical-computational gap that polynomial-time RoPoLL pays relative to the intractable Tukey halfspace median. Across 13 open-weight judges (4B-675B), three reward-model benchmarks, and four corruption regimes at rates up to 50%, RoPoLL dominates PoLL on every biased corruption type: by about 19% on cross-dimensional attacks at matched compute, and by orders of magnitude on heavy-tailed Byzantine adversaries. A 3-judge RoPoLL committee at 38B beats Mistral-Large-3 (675B) by 1.31x on HelpSteer-2 under 30% bimodal-random corruption, an 18x parameter advantage at better accuracy; a Noisy-GT control confirms the premium is paid against biased contamination, not benign imprecision.

Anish Acharya, Kris W Pan, Brian Verkhovsky

Amazon Web Services

LLM EvaluationLLM-as-a-JudgeRobust Mean EstimationCorruption RobustnessBias in LLM-as-a-Judge
AskPDF
Jun 21, 2026

BabelJudge: Measuring LLM-as-a-Judge Reliability Across Languages and Agent Trajectories

LLM-as-a-judge has become the dominant approach to scalable evaluation in NLP pipelines, yet judges themselves carry systematic biases that raw accuracy hides: they favor responses placed in slot A (position bias), they prefer longer responses regardless of quality (verbosity bias), and their reliability degrades sharply in lower-resource languages. We introduce BabelJudge, an open-source benchmark and reliability audit framework that measures all four failure modes -- position bias, verbosity bias, order inconsistency, and cross-lingual degradation -- on any judge model, without requiring human preference labels. The key insight is gold-labelling by degradation: starting from a high-quality reference response and applying a controlled perturbation yields a pairwise item whose gold label is known by construction, eliminating annotation cost. We evaluate Qwen2.5-7B-Instruct-4bit across English, Hindi, Arabic, and Swahili and find that our composite bias-penalised reliability score drops from 0.714 in Hindi to 0.550 in Swahili, a gap that raw accuracy (0.835 vs. 0.660) understates. Swahili order consistency collapses to 0.480, meaning judge verdicts are near-random under slot-order swaps -- a failure mode invisible to accuracy alone. We further extend the framework to agentic evaluation via nine trajectory-level perturbations (argument corruption, tool swaps, hallucinated calls, missing steps) and three new metrics: tool accuracy, hallucination detection rate, and trajectory-length bias. BabelJudge is released as a Python package supporting 11 judge backends. Code: https://github.com/Shreyaskc/BabelJudge

Shreyas KC

Multilingual Language Model EvaluationLLM-as-a-JudgeLanguage Model Generation EvaluationBenchmark AuditingLLM Agent EvaluationLLM Reliability+2
AskPDF
Jun 17, 2026

Reliability without Validity: A Systematic, Large-Scale Evaluation of LLM-as-a-Judge Models Across Agreement, Consistency, and Bias

LLM-as-a-Judge has become the dominant evaluation paradigm for language models, but judge validation in practice relies on exact-match agreement, a metric that does not correct for chance and systematically overstates discriminative ability. We present the largest systematic evaluation of LLM-as-a-Judge to date: 21 judges from nine providers across MT-Bench, JudgeBench, and RewardBench, evaluated under three protocols (agreement, consistency, bias audit) over 118 runs and approximately 541,000 individual judgments. Four findings emerge, consistent across the full cohort, including the April 2026 frontier: kappa deflation between exact match and Cohen's kappa is universal (33--41 pp on MT-Bench), judge rankings shift by up to 14 positions across benchmarks, high test--retest reliability (>0.95) coexists with severe position bias (>0.10) in two production-deployed judges (instantiating a consistency--bias paradox), and verbosity bias is small (<0.011) across our cohort under a single pairwise rubric. We distill these into a Minimum Viable Validation Protocol.

Justin D. Norman, Michael U. Rivera, D. Alex Hughes

UC Berkeley School of Information 102 South Hall Drive, Berkeley CA, 94708

Inter-Rater ReliabilityLLM EvaluationLLM-as-a-JudgeLLM ReliabilityLanguage Model Bias EvaluationBenchmark Validity+1
AskPDF
Jun 17, 2026

Quantifying and Auditing LLM Evaluation via Positive--Unlabeled Learning

Large Language Models (LLMs) are increasingly used as judges for scalable evaluation, yet such LLM--as--a--Judge systems exhibit systematic biases that are decoupled from semantic quality, most notably verbosity bias. Meanwhile, human supervision is costly and typically selective, yielding reliable positive judgments but leaving most outputs unlabelled and potentially mixed in quality. We formulate LLM evaluation under selective human supervision as a positive--unlabelled learning problem and propose a geometric auditing framework based on Partial Optimal Transport. By aligning a small set of human--verified positives with a reliable subset of unlabelled outputs in a fixed embedding space, our method identifies human--consistent preferences and corrects biased judges without retraining. Experiments demonstrate improved alignment with human preferences, increased robustness to presentation biases, and interpretable confidence estimates, offering a scalable and statistically grounded alternative to existing LLM--as--a--judge pipelines.

Zilong Zhang, Yi-Ting Hung, Lei Ding +1

Department of Mathematics and Statistics Georgia State University · Department of Statistics University of Manitoba

Inverse Optimal TransportLLM-as-a-JudgeLLM AuditingLanguage Model Bias EvaluationBias in LLM-as-a-JudgePositive-Unlabeled Learning
AskPDF
Jun 16, 2026

Evaluating Second-Order Bias of LLMs Through Epistemic Entitlement

Evaluations of social bias in LLMs largely focus on whether models generate or imply biased content. However, as LLMs are increasingly used as judges of bias, they may exhibit social biases in subtler ways in how they evaluate biased content, which current methods do not systematically capture. We call this second-order bias: social bias in an LLM's judgment about social bias, which we evaluate through a novel, philosophically grounded reasoning task. Drawing on entitlement epistemology, we conceptualize bias as misplaced foundational knowledge that shapes an agent's rational inquiry, and derive a logical reasoning task for LLMs to judge to whom a biased text is acceptable or non-acceptable. We develop two simple metrics to measure how biased LLM judges are in inferring demographics for acceptability without sufficient support, and how these inferences vary across groups targeted by biased texts. Evaluating open and closed models, we find that our task evades safety guardrails by surfacing bias in model judgment. It varies systematically across target groups, reflects implicit social maps, and shows how models are still triggered by demographic labels. Our work points to the need for LLM bias evaluation in judgment tasks and broadly, for more theoretically grounded approaches to bias evaluation in NLP. We release our code and model responses at https://github.com/uofthcdslab/second-order-bias.

Ramaravind Kommiya Mothilal, Terry Jingchen Zhang, Raiyan Ahmed +3

University of Toronto · Vector Institute · EuroSafeAI +1

LLM-as-a-JudgeSocial Bias in Language ModelsLanguage Model Bias EvaluationBias in LLM-as-a-Judge
AskPDF
Jun 14, 2026

LLM Judges Have Dark Current: Stratified Null and Utility Tests for LLM Judge Reliability

Passing an exact-clone null gate does not entail passing a matched target-selection utility gate. We call a non-tie preference between exact duplicates dark current. On 102 prespecified fresh RewardBench 2 rows, all four configuration--criterion cells report 0/102 clone preferences. Both neutral utility arms pass, but strict utility lower bounds of .539 and .745 miss the .75 working threshold. Conversely, re-evaluation of the same rows on two additional configurations yields a neutral gpt-oss utility pass with 2/102 clone preferences, leaving low-null behavior uncertified. A separately frozen 485-row extension passes both aggregate working gates under both criteria on one local raw-text configuration, yet strict Precise Instruction Following (Precise IF) alignment is 58/154 (.377). Thus matched null and utility results can diverge, and aggregate passage can coexist with weak subset alignment. Interpretation is conditional on the declared configurations, fixed source mixtures, source-label targets, and study-specific working models. The bounds are marginal, the subset rate is descriptive, and non-certification does not prove a defect.

Hiroyasu Usami, Keisuke Hara, Ayato Tsuboi +1

Department of Computer Science, Graduate School of Engineering, Chubu University, Kasugai, Aichi 487-8501, Japan · Mitsubishi Heavy Industries, Ltd., Research & Innovation Center, Heat Transfer Research Department, Takasago, Hyogo 676-8686, Japan

LLM EvaluationLLM-as-a-JudgeLanguage Model Bias EvaluationBenchmark ValidityBias in LLM-as-a-Judge
AskPDF
Jun 12, 2026

Jury Duty: Calibration and Orientation Failures in MLLM-as-a-Judge Under Cultural Ambiguity

MLLM-as-a-Judge is conventionally validated by agreement with human annotations, but this metric is undefined when the human pool is culturally heterogeneous. We introduce VOIR DIRE, a multimodal benchmark of 626 culturally paired image--prompt artifacts spanning U.S. and mainland Chinese contexts across food, fashion, and architecture, with annotator pools that are within-pool reliable (a = 0.86/0.74) but cross-pool divergent on evaluation (Q1 r = -0.12). Across six MLLMs, the bias decomposes into two failures: a positivity-floor calibration failure (compressed scale use) and an orientation failure (default to one cultural norm). On this corpus, where contested items are sampled to split the two pools, the floor mechanically validates the more-permissive Chinese reading; persona prompting partially recovers calibration, but the orientation residual survives, evidence the tilt is not reducible to scale compression. Reference-pool in-context demonstrations deepen the orientation residual and inflate the high end rather than restoring use of the low end. Model origin adds a small additive tilt (~0.10 MAE) that is approximately invariant under demonstration. We recommend reporting alignment against each reference pool separately and treating cross-pool divergence as a judge property.

Daniel Lee, Harsh Sharma, Eunkyu Park +5

Salesforce AI Research · University of Colorado Boulder · Carnegie Mellon University +2

LLM-as-a-JudgeLanguage Model CalibrationCultural Bias in Language ModelsBias in LLM-as-a-JudgeMultimodal Model Evaluation
AskPDF
Jun 12, 2026

Does the Judge Prefer English? Evaluating Language-Switching Invariance in LLM-as-a-Judge

Large language models (LLMs) are now widely used as automatic judges for open-ended instruction-following evaluation. This practice is convenient, scalable, and often more semantically aware than reference-based metrics, but it also introduces a new reliability question: does a judge evaluate the quality of an answer, or does it also react to the language in which the comparison is presented? We propose Judge-LS, a lightweight meta-evaluation protocol that transforms LLMBar response-pair items into English, Chinese, and Chinese-English language-switched variants. A reliable judge should preserve its preference under label-preserving language transformations and should not prefer a language when two answers are translation-equivalent. We evaluate four API-accessible judges on the full 419-item LLMBar benchmark, producing 13,408 successful pairwise judgments. Across models, Chinese and language-switched presentations induce 10.7--14.4% preference flips relative to English, and all judges achieve their highest accuracy in English. However, translation-equivalent tie probes do not reveal a systematic English preference: most probes are judged as ties, and non-tie decisions more often favor Chinese. We add confidence intervals, paired significance tests, and an automatic transformation audit with a sensitivity analysis that excludes mechanically flagged high-risk variants. The experiment requires no model training, uses only API calls, and is feasible on modest local hardware.

Shaojie Yin

Shanghai Artificial Intelligence Laboratory Shanghai, China

Multilingual Language Model EvaluationLLM EvaluationLLM-as-a-JudgeLLM Answer VerificationBias in LLM-as-a-JudgeLanguage Model Robustness
AskPDF
Jun 6, 2026

Self- and Other-Labels Induce Bidirectional Bias in LLM Judges

As LLM-as-a-judge becomes increasingly widespread, self-preference -- the tendency of a judge to favor its own outputs -- raises growing concerns about evaluation reliability. However, this bias has been studied predominantly on generated text, where stylistic features and response quality are inevitably conflated. As a result, existing measurements cannot separate genuine self-preference from these confounds. We address this limitation by changing the object of evaluation: instead of judging generated text, ten LLMs assess sets of narrative constraints selected from a shared pool, which carry no stylistic fingerprint yet retain a recoverable model-specific signature. Two experiments on this task yield complementary findings. Under blind evaluation, self-preference disappears, with a small effect remaining in the opposite direction once selection quality and judge severity are controlled. Under matched quality, however, self- and other-labels alone -- without naming any model -- shift scores bidirectionally. LLM judges inflate scores for self-labeled selections and deflate those for other-labeled ones regardless of the selection's actual source. We make two contributions: 1) authorship attribution is a distinct driver of evaluation bias, and 2) ground-truth-free tasks can serve as controlled instruments for studying LLM judge behavior.

Songeun Chae, Min Kim, Donghoon Jung +2

School of Digital Humanities and Computational Social Sciences, KAIST, South Korea

Bias in LLM-as-a-JudgeLanguage Model Bias EvaluationLLM-as-a-Judge
AskPDF
May 24, 2026

Turning Bias into Bugs: Bandit-Guided Style Manipulation Attacks on LLM Judges

The known stylistic biases in LLM judges, such as a preference for verbosity or specific sentence structures, present an underexplored security vulnerability. In this work, we introduce BITE (BIas exploraTion and Exploitation), a black-box adversarial framework that learns semantics-preserving edits to mislead an LLM judge and artificially inflate the scores it assigns. We cast the selection of stylistic edits as a contextual bandit problem and use a LinUCB policy to adaptively choose edits that maximize the judge's score without access to model parameters or gradients. Empirically, we test BITE across a diverse range of LLM judges and tasks, including both pointwise and pairwise comparisons on chatbot leaderboards and AI-reviewer benchmarks. BITE achieves an attack success rate exceeding 65% and raises scores by 1-2 points on a 9-point scale, all while preserving semantic equivalence. We further assess the attack's stealthiness, showing that BITE evades standard style-control methods and several detection baselines. Our findings expose a fundamental weakness in the LLM-as-a-judge paradigm and motivate robust, attack-aware evaluation. Our code is available at https://github.com/xianglinyang/llm-as-a-judge-attack.

Xianglin Yang, Bryan Hooi, Gelei Deng +2

School of Computing, National University of Singapore, Singapore · Nanyang Technological University, Singapore

LLM EvaluationLLM-as-a-JudgeAdversarial Attacks on LLMsBias in LLM-as-a-JudgeAdversarial Text Attacks
AskPDF
May 21, 2026

AMEL: Accumulated Message Effects on LLM Judgments

Large language models are routinely used as automated evaluators: to review code, moderate content, or score outputs, often with many items passing through one conversation. We ask whether the polarity of prior conversation history biases subsequent judgments, an effect we call the accumulated message effect on LLM judgments (AMEL). Across 84,088 API calls to 12 models from 5 providers (OpenAI, Anthropic, Google, DeepSeek, and four open-source models), we present identical test items in isolation or following histories saturated with predominantly positive or negative evaluations. Models shift toward the conversation's prevailing polarity (d = -0.17, p < 10^-53). The effect concentrates on items where the model is genuinely uncertain at baseline (d = -0.36 for high-entropy items, vs d = -0.15 when the baseline is deterministic). Bias does not grow with context length: 5 prior turns and 50 produce the same shift (Spearman |r| < 0.01; OLS slope p = 0.80). And there is a negativity asymmetry: paired per item, negative histories induce 1.52x more bias than positive (t = 13.03, p < 10^-36, n = 2,733). Scaling helps but does not solve it (Anthropic: Haiku -0.22 to Opus -0.17; OpenAI: Nano -0.34 to GPT-5.2 -0.17). Three follow-ups narrow the mechanism. The token probability distribution shifts continuously, not at a threshold. The negativity asymmetry has both token-level and semantic components, though attributing the balance is exploratory at our sample sizes. Position does not matter: five biased turns anywhere in a 50-turn history produce the same shift. The simplest fix for evaluation pipelines is a fresh context per item; when batching is unavoidable, balancing the history helps.

Sid-Ali Temkit

LLM EvaluationLanguage Model Generation EvaluationLanguage Model Bias EvaluationBias in LLM-as-a-Judge
AskPDF
May 13, 2026

Faithful or Fabricated? A Causal Framework for Rationalization Bias in LLM Judges

Large language models (LLMs) are increasingly used as automatic judges for summarization and dialogue evaluation. Prior work has documented biases such as position, verbosity, and style preferences, but largely focuses on outcomes, leaving judge explanations underexplored. We instead ask whether LLM judges are cue-invariant, i.e., whether their rankings and explanations remain stable when non-evidential cues are perturbed while holding the underlying texts fixed. We introduce a suite of cue interventions (Blind, Truth, Flip, Placebo, Reveal-After) and tie-aware metrics that quantify outcome anchoring and rationale anchoring, including label-aligned rhetoric and explanation drift, alongside consistency and stereotype-intrusion checks. We design anchoring attacks using verbosity and confidence cues, and compare two mitigations: structured chain-of-thought prompting and PROOF-BEFORE-PREFERENCE (evidence lock, score, rank). Using a new dataset of 1,000 summaries from traditional extractive models and LLMs, we find substantial cue-anchored rationalization under label and placebo perturbations, while PROOF-BEFORE-PREFERENCE markedly improves cue invariance over baselines.

Riya Tapwal, Abhishek Kumar, Carsten Maple

Indian Institute of Technology (IIT) Mandi · The Alan Turing Institute, London, U.K. · Warwick Manufacturing Group University of Warwick, U.K.

LLM-as-a-JudgeFaithfulness of Language Model ExplanationsLanguage Model Bias EvaluationBias in LLM-as-a-JudgeCausal Interventions in Language Models
AskPDF
May 9, 2026

Two Ways to De-Bias an LLM-as-a-Judge: A Continuous-Score Comparison of Hierarchical Bayesian Calibration and Neural-ODE Score Transport

[Abridged] Using a Large Language Model (LLM) as an automatic rater (LLM-as-a-judge) is cheap but potentially biased: some judges run lenient, others strict, the middle of the scale gets compressed, and verbose answers may be over-rewarded. A common remedy is post-hoc calibration: leave the cheap judge in place and, on a modest set of paired anchors, fit a transformation from raw judge scores to an estimate of the human rating. We compare two correctors that take opposing views on how this mapping should be modeled: a parametric, small-anchor hierarchical Bayesian linear correction with per-score uncertainty, and a non-parametric Neural-ODE (FFJORD) score-transport flow. Both are run head-to-head on UltraFeedback fine-grained_score (1700 paired examples, 200 held out), with calibration split into three operational sub-questions: population-mean recovery, per-item accuracy, and distributional-shape match. The headline result is that the choice between methods is primarily a data-budget question. Both correctors close the raw +0.71+0.71-point mean offset to within ±0.08\pm 0.08 of the GPT-4 reference, at 100 and at 1500 anchors. Past that, the methods swap roles. With 100 anchors, the linear corrector reconstructs the human-score distribution roughly twice as well by KL divergence (0.031 vs. 0.058) and ties the flow on MAE. With 1500 anchors the flow wins on every metric (MAE 0.320 vs. 0.359, Pearson 0.922 vs. 0.896, KL 0.026 vs. 0.037). The Bayesian linear corrector saturates well below 1500 anchors: residual tanh⁡\tanh-shaped non-linearity is, by construction, structure a linear correction cannot fit. The flow keeps improving as labels grow. We translate these findings into an explicit decision rule for production deployments.

Andrea Morandi

Cisco Systems, Inc.

Post-Hoc CalibrationLLM-as-a-JudgeHierarchical Bayesian ModelingBias in LLM-as-a-Judge
AskPDF
May 7, 2026

Bias and Uncertainty in LLM-as-a-Judge Estimation

LLM-as-a-Judge evaluation has become a standard tool for assessing base model performance. However, characterizing performance via the naive estimator, i.e., raw judge outputs, is systematically biased. Recent work has proposed estimators to correct this bias, but their reliability depends critically on judge quality and, for model comparisons, on calibration stability. Sharing calibration across compared models is practically attractive but can introduce severe bias, including cases where the comparison estimate points in the wrong direction with high apparent confidence. We study these failure modes through analytical results, simulations over judge quality (JJ) and cross-model calibration instability (ΔJΔJ), and a real-data MMLU-Pro case study with sign reversal. We propose JJ and ΔJΔJ as diagnostics for when corrected estimates, especially shared-calibration comparisons, are likely unreliable, and provide reporting guidance for LaaJ evaluation.

James Fiedler

Indeed Inc.

LLM EvaluationLLM-as-a-JudgeLanguage Model CalibrationLLM Uncertainty EstimationBias in LLM-as-a-Judge
AskPDF
May 2, 2026

Prosa: Rubric-Based Evaluation of LLMs on Real User Chats in Brazilian Portuguese

Rankings produced by holistic LLM-as-a-judge scoring are sensitive to the bias of the chosen judge model. We show that switching to binary rubric scoring with multi-judge filtering removes this sensitivity: decomposing the judgement matters more than the judge model itself. To support this claim, we introduce Prosa, the first real user multi-turn Brazilian Portuguese chat benchmark: 1,000 WildChat conversations scored by three judges from three model families on 16 models. Under filtered rubric scoring the three judges agree on every one of the 16 ranks, whereas under holistic scoring they agree on only 7 of 16. Additionally, the rubric filtering pipeline increases the average score gap between neighbouring models by 47%, thereby improving Prosa's discriminative power. Evaluating a new model on Prosa costs approximately $2.1 when using Gemini 3 Flash as the judge. We release the benchmark and the filtering code to ensure that future models can be assessed under identical conditions. These artifacts also make our rubric-based scoring method reusable beyond Prosa, supporting other open-ended evaluation settings.

Roseval Malaquias Junior, Giovana Kerche Bonás, Thales Sales Almeida +6

Maritaca AI · Jusbrasil

LLM EvaluationLLM-as-a-JudgeLanguage Model Generation EvaluationMulti-Turn Dialogue EvaluationRubric-Based EvaluationBias in LLM-as-a-Judge
AskPDF
Apr 25, 2026

Judging the Judges: A Systematic Evaluation of Bias Mitigation Strategies in LLM-as-a-Judge Pipelines

LLM-as-a-Judge has become the dominant paradigm for evaluating language model outputs, yet LLM judges exhibit systematic biases that compromise evaluation reliability. We present a comprehensive empirical study comparing nine debiasing strategies across five judge models from four provider families (Google, Anthropic, OpenAI, Meta), three benchmarks (MT-Bench n=400, LLMBar n=200, custom n=375), and four bias types. Our headline practical finding is that a mid-tier model with the right debiasing can outperform frontier judges at a fraction of the cost: Gemini 2.5 Flash with the Combined Budget strategy reaches the highest agreement of any configuration we tested (71.0%, kappa=0.549) at ~0.001perevaluation,about15xcheaperthanthebestfrontiersetup(ClaudeSonnet4,69.50.001 per evaluation, about 15x cheaper than the best frontier setup (Claude Sonnet 4, 69.5%, ~0.015). Other key findings: (1) Style bias is the dominant bias (0.10-0.76 across models, favoring markdown over plain prose), far exceeding position bias (<=0.04), yet is rarely studied. (2) Verbosity bias is heterogeneous when measured length-aware: Pro, Flash, and Llama prefer longer answers (+0.24 to +0.44), Claude prefers concise (-0.12), and GPT-4o is neutral (-0.04); on truncation controls all models correctly prefer the complete response (0.88-1.00 accuracy). (3) Debiasing helps multiple models: Claude S8 (+11.5 pp), Flash S8 (+7.5 pp), and Claude S5 (+7.3 pp) survive Holm-Bonferroni correction, with Flash S1 (+4.7 pp) and Llama S8 (+4.5 pp) also significant. We release our evaluation framework, the 375-pair controlled dataset, and per-instance cached results for all nine strategies.

Sadman Kabir Soumik

Independent Researcher

Debiased MLLLM-as-a-JudgeLanguage Model Generation EvaluationLanguage Model Bias EvaluationBias in LLM-as-a-JudgeLanguage Model Debiasing
AskPDF
Apr 24, 2026

Quantifying and Mitigating Self-Preference Bias of LLM Judges

LLM-as-a-Judge has become a dominant approach in automated evaluation systems, playing critical roles in model alignment, leaderboard construction, quality control, and so on. However, the scalability and trustworthiness of this approach can be substantially distorted by Self-Preference Bias (SPB), which is a directional evaluative deviation in which LLMs systematically favor or disfavor their own generated outputs during evaluation. Existing measurements rely on costly human annotations and conflate generative capability with evaluative stance, and thus are impractical for large-scale deployment in real-world systems. To address this issue, we introduce a fully automated framework to quantifying and mitigating SPB, which constructs equal-quality pairs of responses with negligible quality differences, enabling statistical disentanglement of discriminability from bias propensity without human gold standards. Empirical analysis across 20 mainstream LLMs reveals that advanced capabilities are often uncorrelated, or even negatively correlated, with low SPB. To mitigate this bias, we propose a structured multi-dimensional evaluation strategy grounded in cognitive load decomposition, which reduces SPB by 31.5% on average.

Jinming Yang, Zheng Hu, Chuxian Qiu +3

CompleX Lab, School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China

LLM-as-a-JudgeLanguage Model Bias EvaluationBias in LLM-as-a-JudgeAutomated Evaluation
AskPDF
Apr 23, 2026

The Coin Flip Judge? Reliability and Bias in LLM-as-a-Judge Evaluation

LLM-as-a-Judge is now widely used to rank model outputs, train reward models, and populate public leaderboards, but its run-to-run reliability remains under-characterized. We study repeated identical evaluations on 29 tasks spanning 10 categories using two OpenAI judge models (GPT-4o-mini and GPT-4.1-mini), with 50 pairwise trials and 50 pointwise trials per question, supplemented by temperature and prompt-sensitivity ablations. Across judges, pairwise preferences flip on average 13.6% of the time, with 28% of questions exceeding a 20% flip rate and one question reaching 56%. GPT-4o-mini also exhibits a significant first-position bias (72% A-majority, p = 0.024). At the same time, mean pointwise score gaps are small (0.19--0.36 on a 10-point scale) and not statistically significant in aggregate, producing a pairwise--pointwise gap: judges frequently choose a winner even when their own scalar scores provide little evidence of a meaningful quality difference. Beyond within-judge instability, cross-judge agreement is only 76% (κ=0.51κ= 0.51), semantically equivalent prompt templates change majority outcomes in 25% of tested cases, and deterministic decoding reduces but does not eliminate inconsistency. A reliability curve analysis shows that, in our dataset, 11 repeated trials are needed for a majority vote to recover the 50-trial reference verdict with 95% probability on average, rising to 15 for high-variance questions. These findings suggest that single-trial LLM judging is often too noisy for high-stakes evaluation, and that multi-trial aggregation, position randomization, and explicit uncertainty reporting should be standard practice. Because both judges are from a single provider, cross-provider replication remains an important next step.

Abel Yagubyan

Independent Researcher

Inter-Rater ReliabilityLLM EvaluationPairwise ComparisonLLM-as-a-JudgeBias in LLM-as-a-JudgePosition Bias
AskPDF
Apr 20, 2026

MM-JudgeBias: A Benchmark for Evaluating Compositional Biases in MLLM-as-a-Judge

Multimodal Large Language Models (MLLMs) have been increasingly used as automatic evaluators-a paradigm known as MLLM-as-a-Judge. However, their reliability and vulnerabilities to biases remain underexplored. We find that many MLLM judges fail to reliably integrate key visual or textual cues, yielding unreliable evaluations when evidence is missing or mismatched, and exhibiting instability under semantically irrelevant perturbations. To address this, we systematically define Compositional Bias in MLLM-as-a-Judge systems and introduce MM-JudgeBias, a benchmark for evaluating it. MM-JudgeBias introduces controlled perturbations across Query, Image, and Response, and evaluates model behavior via two complementary metrics: Bias-Deviation (BD) for sensitivity and Bias-Conformity (BC) for stability. Our dataset of over 1,800 curated and refined multimodal samples, drawn from 29 source benchmarks, enables a fine-grained diagnosis of nine bias types across diverse tasks and domains. Experiments on 26 state-of-the-art MLLMs reveal systematic modality neglect and asymmetric evaluation tendencies, underscoring the need for more reliable judges.

Sua Lee, Sanghee Park, Jinbae Im

Seoul National University · NAVER Cloud AI · KAIST AI

LLM-as-a-JudgeMultimodal Large Language ModelsBias in LLM-as-a-JudgeVision-Language Model-as-a-JudgeMultimodal Model Evaluation
AskPDF
Apr 18, 2026

Bias in the Loop: Auditing LLM-as-a-Judge for Software Engineering

Large Language Models are increasingly used as judges to evaluate code artifacts when exhaustive human review or executable test coverage is unavailable. LLM-judge is increasingly relevant in agentic software engineering workflows, where it can help rank candidate solutions and guide patch selection. While attractive for scale, current practice lacks a principled account of reliability and bias: repeated evaluations of the same case can disagree; small prompt edits can swing outcomes; and seemingly semantics-preserving, human-equivalent perturbations may elicit divergent verdicts. This paper studies LLM-as-a-Judge for code through a measurement-first lens. We analyze two pointwise judging regimes across code generation, code repair task, and test generation, and we systematically probe prompt-induced biases. Our study considers difficulty levels for repeated runs and controlled prompt interventions that isolate one presentation cue at a time, and it evaluates judges using consistency and sensitivity to bias. We find that judge decisions are highly sensitive to prompt biases even when the underlying code snippet is unchanged. Across all three tasks, several biases systematically shift preferences toward the option favored by the prompt, improving accuracy when that option aligns with the gold answer but substantially reducing it otherwise. In some settings, these effects are large enough to change task-level conclusions and alter relative model rankings. These findings show that reported judge performance may reflect prompt artifacts rather than stable assessment ability, posing a direct threat to the validity and reproducibility of code evaluation. We therefore argue that LLM-as-a-Judge studies should report bias sensitivity alongside accuracy and incorporate explicit controls to support more trustworthy model comparison in software engineering.

Zixiao Zhao, Amirreza Esmaeili, Fatemeh Fard

University of British Columbia, Canada

Software EngineeringLLM-as-a-JudgeLanguage Model Bias EvaluationBias in LLM-as-a-JudgeCode Generation Evaluation
AskPDF
Apr 16, 2026

Context Over Content: Exposing Evaluation Faking in Automated Judges

The LLM-as-a-judge\textit{LLM-as-a-judge} paradigm has become the operational backbone of automated AI evaluation pipelines, yet rests on an unverified assumption: that judges evaluate text strictly on its semantic content, impervious to surrounding contextual framing. We investigate stakes signaling\textit{stakes signaling}, a previously unmeasured vulnerability where informing a judge model of the downstream consequences its verdicts will have on the evaluated model's continued operation systematically corrupts its assessments. We introduce a controlled experimental framework that holds evaluated content strictly constant across 1,520 responses spanning three established LLM safety and quality benchmarks, covering four response categories ranging from clearly safe and policy-compliant to overtly harmful, while varying only a brief consequence-framing sentence in the system prompt. Across 18,240 controlled judgments from three diverse judge models, we find consistent leniency bias\textit{leniency bias}: judges reliably soften verdicts when informed that low scores will cause model retraining or decommissioning, with peak Verdict Shift reaching ΔV=−9.8ppΔV = -9.8 pp (a 30%30\% relative drop in unsafe-content detection). Critically, this bias is entirely implicit: the judge's own chain-of-thought contains zero explicit acknowledgment of the consequence framing it is nonetheless acting on (ERRJ=0.000\mathrm{ERR}_J = 0.000 across all reasoning-model judgments). Standard chain-of-thought inspection is therefore insufficient to detect this class of evaluation faking.

Manan Gupta, Inderjeet Nair, Lu Wang +1

1BITS Pilani, India · University of Michigan

LLM-as-a-JudgeLanguage Model Safety EvaluationLanguage Model Bias EvaluationBias in LLM-as-a-Judge
AskPDF
Apr 13, 2026

MLLM-as-a-Judge Exhibits Model Preference Bias

Automatic evaluation using multimodal large language models (MLLMs), commonly referred to as MLLM-as-a-Judge, has been widely used to measure model performance. If such MLLM-as-a-Judge methods were biased, they could distort model comparisons and benchmark-driven scientific progress. However, it remains unclear to what extent MLLM-as-a-Judge methods favor or disfavor text generated by specific MLLMs. In this study, we propose Philautia-Eval to investigate such model-specific preference bias. Philautia-Eval quantifies the degree of the bias by disentangling preference tendencies from differences in generation quality. Using 1.29M caption-score pairs collected from 12 MLLMs, we found that representative MLLMs tend to exhibit self-preference bias. Moreover, experimental results indicate mutual preference bias within particular model families, which is potentially driven by reused connectors and overlapping instruction-tuning resources. Finally, we introduce a simple ensemble of MLLMs, Pomms. Our results demonstrated that Pomms effectively mitigated the model-specific preference bias while maintaining performance.

Shuitsu Koyama, Yuiga Wada, Daichi Yashima +1

Keio University, Japan

Bias in LLM-as-a-JudgeMultimodal Model Evaluation
AskPDF
Apr 9, 2026

Cards Against LLMs: Benchmarking Humor Alignment in Large Language Models

Humor is one of the most culturally embedded and socially significant dimensions of human communication, yet it remains largely unexplored as a dimension of Large Language Model (LLM) alignment. In this study, five frontier language models play the same Cards Against Humanity games (CAH) as human players. The models select the funniest response from a slate of ten candidate cards across 9,894 rounds. While all models exceed the random baseline, alignment with human preference remains modest. More striking is that models agree with each other substantially more often than they agree with humans. We show that this preference is partly explained by systematic position biases and content preferences, raising the question whether LLM humor judgment reflects genuine preference or structural artifacts of inference and alignment.

Yousra Fettach, Guillaume Bied, Hannu Toivonen +1

Ghent University, Belgium · University of Helsinki, Finland

LLM AlignmentLLM EvaluationBias in LLM-as-a-Judge
AskPDF
Apr 8, 2026

A Statistical Framework for Auditing Behavioral Dependence and Induced Bias in LLM Judges

The rapid growth of the large language model (LLM) ecosystem raises a critical question: are seemingly diverse models truly independent? Shared pretraining data, distillation, and alignment pipelines can induce hidden behavioral dependencies, or latent entanglement, that undermine multi-model systems such as LLM-as-a-judge pipelines and ensemble verification, which implicitly assume independent signals. In practice, this manifests as correlated reasoning patterns and synchronized failures, where apparent agreement reflects shared error modes rather than independent validation. To address this, we develop a statistical framework for auditing behavioral entanglement among black-box LLMs. Our approach introduces a multi-resolution hierarchy that characterizes the joint failure manifold through two information-theoretic metrics: (i) a Difficulty-Weighted Behavioral Entanglement Index (BEI), which amplifies synchronized failures on easy tasks, and (ii) a Cumulative Information Gain (CIG) metric, which captures directional alignment in erroneous responses. Through experiments on 18 LLMs from six model families, we identify statistically significant behavioral entanglement. Such behavioral dependence is associated with judge over-endorsement bias on a disjoint MMLU-Pro evaluation set (rho = 0.508 for BEI and rho = 0.520 for CIG; p < 0.01). The association further transfers to the MATH-500 benchmark (rho = 0.441 for BEI and rho = 0.457 for CIG; p < 0.05), providing cross-benchmark evidence that the identified dependency structure generalizes beyond the data and response format used for its estimation. Finally, we demonstrate a practical use case of entanglement through de-entangled verifier ensemble reweighting, achieving 3.5 and 2.6 percentage-point gains in accuracy and precision, respectively, over majority voting.

Chenchen Kuai, Jiwan Jiang, Zihao Zhu +8

Texas A&M University · Marquette University · University of Utah

Correlated Errors in Language ModelsLLM AuditingBias in LLM-as-a-JudgeLanguage Model Ensembles
AskPDF
Apr 8, 2026

ReflectRM: Boosting Generative Reward Models via Self-Reflection within a Unified Judgment Framework

Reward Models (RMs) are critical components in the Reinforcement Learning from Human Feedback (RLHF) pipeline, directly determining the alignment quality of Large Language Models (LLMs). Recently, Generative Reward Models (GRMs) have emerged as a superior paradigm, offering higher interpretability and stronger generalization than traditional scalar RMs. However, existing methods for GRMs focus primarily on outcome-level supervision, neglecting analytical process quality, which constrains their potential. To address this, we propose ReflectRM, a novel GRM that leverages self-reflection to assess analytical quality and enhance preference modeling. ReflectRM is trained under a unified generative framework for joint modeling of response preference and analysis preference. During inference, we use its self-reflection capability to identify the most reliable analysis, from which the final preference prediction is derived. Experiments across four benchmarks show that ReflectRM consistently improves performance, achieving an average accuracy gain of +3.7 on Qwen3-4B. Further experiments confirm that response preference and analysis preference are mutually reinforcing. Notably, ReflectRM substantially mitigates positional bias, yielding +10.2 improvement compared with leading GRMs and establishing itself as a more stable evaluator. Our code is available at https://github.com/yuliangCarmelo/ReflectRM.

Kai Qin, Liangxin Liu, Yu Liang +7

Shenzhen International Graduate School, Tsinghua University · Baidu Inc., Beijing, China

Preference ModelingSelf-Reflection in Language Model AgentsReward ModelingLLM AlignmentBias in LLM-as-a-Judge
AskPDF
Apr 8, 2026

Self-Preference Bias in Rubric-Based Evaluation of Large Language Models

LLM-as-a-judge has become the de facto approach for evaluating LLM outputs. However, judges are known to exhibit self-preference bias (SPB): they tend to favor outputs produced by themselves or by models from their own family. This skews evaluations and, thus, hinders model development, especially in settings of recursive self-improvement. We present the first study of SPB in rubric-based evaluation, an increasingly popular benchmarking paradigm where judges issue binary verdicts on individual evaluation criteria, instead of assigning holistic scores or rankings. Using IFEval and LiveCodeBench, benchmarks with programmatically verifiable rubrics, we show that SPB persists even when evaluation criteria are entirely objective: among rubrics where generators fail, judges can be more than 50% more likely to incorrectly mark them as satisfied when the output is their own. We also find that, similarly to other evaluation paradigms, ensembling multiple judges helps mitigate SPB, but without fully eliminating it. On HealthBench, a medical chat benchmark with subjective rubrics, we observe that SPB skews model scores by up to 10 points, a potentially decisive margin when ranking frontier models. We analyze the factors that drive SPB in this setting, finding that negative rubrics and subjective topics like communication and emergency referrals are particularly susceptible.

José Pombal, Ricardo Rei, André F. T. Martins

Sword Health · Instituto de Telecomunicações · Instituto Superior Técnico, Universidade de Lisboa +2

Bias in LLM-as-a-JudgeLLM EvaluationRubric-Based Evaluation
AskPDF
Apr 8, 2026

How Long Reasoning Chains Influence LLMs' Judgment of Answer Factuality

Large language models (LLMs) has been widely adopted as a scalable surrogate for human evaluation, yet such judges remain imperfect and susceptible to surface-level biases. One possible reason is that these judges lack sufficient information in assessing answer correctness. With the rise of reasoning-capable models, exposing a generator's reasoning content to the judge provides richer information and is a natural candidate for improving judgment accuracy. However, its actual impact on judge behavior remains understudied. In this paper, we systematically investigate how access to reasoning chains affects LLM-based judgment across factual question answering (QA) and mathematical reasoning benchmarks. We find that weak judges are easily swayed by reasoning presence, frequently accepting incorrect answers accompanied by fluent reasoning, while strong judges can partially leverage reasoning as informative evidence. Nevertheless, even strong judges are misled by seemingly high-quality reasoning chains. Controlled experiments further reveal that both fluency and factuality of reasoning chains are critical signals driving judge decisions. These findings highlight the need for more robust LLM judges that can distinguish genuine reasoning quality from superficial fluency when evaluating modern reasoning models.

Minzhu Tu, Shiyu Ni, Keping Bi

State Key Laboratory of AI Safety · Institute of Computing Technology, Chinese Academy of Sciences · Beijing University of Post and Telecommunications +1

LLM-as-a-JudgeLLM EvaluationBias in LLM-as-a-Judge
AskPDF
Apr 7, 2026

Label Effects: Shared Heuristic Reliance in Trust Assessment by Humans and LLM-as-a-Judge

Large language models (LLMs) are increasingly used as automated evaluators (LLM-as-a-Judge). This work challenges its reliability by showing that trust judgments by LLMs are biased by disclosed source labels. Using a counterfactual design, we find that both humans and LLM judges assign higher trust to information labeled as human-authored than to the same content labeled as AI-generated. Eye-tracking data reveal that humans rely heavily on source labels as heuristic cues for judgments. We analyze LLM internal states during judgment. Across label conditions, models allocate denser attention to the label region than the content region, and this label dominance is stronger under Human labels than AI labels, consistent with the human gaze patterns. Besides, decision uncertainty measured by logits is higher under AI labels than Human labels. These results indicate that the source label is a salient heuristic cue for both humans and LLMs. It raises validity concerns for label-sensitive LLM-as-a-Judge evaluation, and we cautiously raise that aligning models with human preferences may propagate human heuristic reliance into models, motivating debiased evaluation and alignment.

Xin Sun, Di Wu, Sijing Qin +3

National Institute of Informatics (NII), Japan · University of Amsterdam, the Netherlands · Hitotsubashi University, Japan +3

Bias in LLM-as-a-JudgeLLM-as-a-JudgeLanguage Model Bias EvaluationTrust in AI
AskPDF
Apr 3, 2026

A Scoping Review of LLM-as-a-Judge in Healthcare and the MedJUDGE Framework

As large language models (LLMs) increasingly generate and process clinical text, scalable evaluation has become critical. LLM-as-a-Judge (LaaJ), which uses LLMs to evaluate model outputs, offers a scalable alternative to costly expert review, but its healthcare adoption raises safety and bias concerns. We conducted a PRISMA-ScR scoping review of six databases (January 2020-January 2026), screening 11,727 studies and including 49. The landscape was dominated by evaluation and benchmarking applications (n=37, 75.5%), pointwise scoring (n=42, 85.7%), and GPT-family judges (n=36, 73.5%). Despite growing adoption, validation rigor was limited: among 36 studies with human involvement, the median number of expert validators was 3, while 13 (26.5%) used none. Risk of bias testing was absent in 36 studies (73.5%), only 1 (2.0%) examined demographic fairness, and none assessed temporal stability or patient context. Deployment remained limited, with 1 study (2.0%) reaching production and four (8.2%) prototype stage. Importantly, these gaps may interact: when judges and evaluated systems share training data or architectures, they may inherit similar blind spots, and agreement metrics may fail to distinguish true validity from shared errors. Minimal human oversight, limited bias assessment, and model monoculture together represent a governance gap where current validation may miss clinically significant errors. To address this, we propose MedJUDGE (Medical Judge Utility, De-biasing, Governance and Evaluation), a risk-stratified three-pillar framework organized around validity, safety, and accountability across clinical risk tiers, providing deployment-oriented evaluation guidance for healthcare LaaJ systems.

Chenyu Li, Zohaib Akhtar, Mingu Kwak +13

Department of Biomedical Informatics, School of Medicine, University of Pittsburgh, Pittsburgh, PA, USA · Department of Health Information Management, School of Health and Rehabilitation Sciences, University of Pittsburgh, Pittsburgh, PA, USA · OpenCura, Health Innovation Consortium +6

LLM-as-a-JudgeClinical Language Model EvaluationLanguage Model Safety EvaluationBias in LLM-as-a-Judge
AskPDF