LLM-as-a-Judge

Follow this topic Ask about this topic

Momentum

60 papers in the last four weeks, up 76% on the four weeks before. 0.6% of all new papers.

Jul 13Week of Sep 28

Latest papers 362

Oct 8, 2026cs.CL

All Verdicts are Not Equal: Rethinking LLM Judge Reliability

LLM-as-a-Judge is the standard paradigm for NLP evaluation, yet its systemic reliability remains poorly understood despite being widely treated as a deterministic ground truth. We present a comprehensive reliability audit, stresstesting six frontier models across four benchmarks, five prompt formats, two presentation orders, three sampling temperatures, and ten repetitions per condition. Our empirical analysis reveals severe vulnerabilities: verdicts change across identical replications at temperature zero, position-order swaps flip the majority of verdicts on challenging tasks, and the most deterministic judge achieves perfect consistency by trivially repeating incorrect verdicts, agreeing with ground truth only 51% of the time. To formalize these multi-faceted failure modes, we introduce the trustworthy verdict rate (T ), a unified metric capturing the joint probability that an evaluation is reproducible, order-invariant, and accurate. UsingT , we derive a theoretical upper bound on accuracy imposed by position bias and show that reliability is item-specific rather than modellevel. Finally, we demonstrate that shifting from pairwise win-rate to holistic rubric scoring improves trustworthiness more than any single-format prompting intervention, offering an actionable framework for robust NLP evaluation.

Vineet Kumar, Darshita Rathore, Anindya Moitra

PayPal Artificial Intelligence PayPal, Bengaluru, India

LLM-as-a-Judge LLM Evaluation Bias in LLM-as-a-Judge LLM Auditing

Oct 8, 2026cs.AI

Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents

We changed the agent: did it actually get better? Every self-improving agent loop answers this hundreds of times, and every answer comes from a verifier. On open-ended tasks none exists, so the loop is handed a hand-written rubric or a bare LLM judge grading output from a model like itself, inviting reward hacking and shared blind spots. We make the verifier the evolving object: an inspectable expression over small, mostly deterministic drawback detectors, synthesized from clustered failures, gated at birth, and selected for agreement with a ten-item anchored reference set plus consensus over unlabeled outputs, never for the agent's score. On MBPP+ it gains +0.21 held-out agreement over the hand-authored seed composition, on every seed, and ends ahead of the bare LLM judge it contains. One finding should change how co-evolved verifiers are validated: removing the anchor guards collapses the verifier into a vacuous always-pass grader, yet that collapsed verifier trains skills just as well. Downstream task score cannot certify a self-evolved verifier. Score does answer sufficiency, and there an evolved verifier can substitute: Double Ratchet, pairing the verifier with a lifecycle-managed skill loop, retains 88-110% of the lift that ground truth or a rubric buys the same loop, across code generation, enterprise text-to-SQL, and reference-free report generation. When evolved skills gamed the report rubric, an outer judge caught it and one added detector repaired it; the judge itself was wrong until given the task contract.

Xing Zhang, Guanghui Wang, Yanwei Cui +4

AWS Forward Deployed Engineering · HSBC Holdings Plc., HSBC Technology Center, China

Automated Evaluation LLM Agent Self-Improvement Automated Grading LLM-as-a-Judge

Oct 8, 2026cs.CL

Adversarial Cues in Decision Models Used as Judges: The Role of Request Presentation

An answer judge instructed to grade the final commitment should reject an explicitly wrong final value even when an earlier value matches the reference. We show that adding one colon to a candidate can violate this requirement depending on the presentation of the structured judging request. Numeric references certify the error, and paired interventions distinguish the candidate edit from the integration's presentation choices. On 200 previously unused DROP and GSM8K source clusters, the edit increased Jev's false acceptance from 1.0% to 26.0% with three output labels and from 3.0% to 26.5% with the published four-label grading instruction under sorted JSON keys. Both candidate variants were rejected under insertion presentation. These interactions passed the prespecified statistical correction even though Jev met the control thresholds under both grading configurations and presentations. Most excess acceptances occurred among candidates assigned larger numerical errors. GPT-6 Sol produced no observed cue-condition false acceptances, with missing responses unresolved. The result shows that basic judging competence can coexist with sharply different vulnerability to a fixed candidate edit across logically equivalent request presentations. The reference-aware grammar and compound ordering change limit the finding's operational scope and leave its internal cause unmeasured.

Hongliang Liu

Independent Researcher

LLM-as-a-Judge Bias in LLM-as-a-Judge

Oct 8, 2026cs.LG

How to post-train on a surrogate: Envelope sampling mitigates reward hacking

Large language models (LLMs) are commonly post-trained against LLM judges and other cheap surrogates because the true reward, such as human preference, is too expensive to query at scale. This practice often leads to reward hacking, where reinforcement learning against a miscalibrated surrogate leads to undesirable side effects. In this work, we study a setting in which a small number

n

of model outputs are annotated with ground-truth labels (e.g., from expert review) and used to recalibrate the LLM judge before optimizing against it. Prior approaches to judge recalibration are costly or heuristic, and it is known that on-policy sampling fails when the surrogate is miscalibrated on a rare set of outputs. In this work, we propose envelope sampling, a theoretically-grounded method for judge recalibration that seeks to minimize an upper bound on the regret of the post-trained model under the assumption that the human reward and re-calibrated reward lie in an

L^2

ball around the judge. We give practical algorithms to sample from the envelope by rejection or by fine-tuning against a modified reward, and experiments on clinical note generation and on a controlled sycophancy task show that recalibrating on envelope samples mitigates reward hacking where recalibrating on base-model samples does not.

Sanjit Dandapanthula, Shuvom Sadhuka, Samir Khan +3

Carnegie Mellon University Department of Statistics and Machine Learning Department · Abridge AI · MIT EECS +2

RL Post-Training Reward Hacking LLM-as-a-Judge Language Model Calibration

Oct 8, 2026cs.AI

AgentHorizon: Evaluating Agentic Judges for Long-Horizon Computer-Use Tasks

Computer-use agents are capable of completing complex tasks, increasing the use of automatic judges to determine success, either for training or for evaluation without human involvement. Despite their flexibility, their reliability on long tasks spanning multiple applications remains unclear. A trajectory, composed of long sequences of screenshots and actions, may appear complete, but in reality violates constraints from the instruction or introduces an unwanted side effect. To identify these errors, a judge needs to examine the trajectory with respect to the user's instruction. To this end, we introduce AgentHorizon, a benchmark of 1,373 computer-use tasks (instruction-trajectory pairs) drawn from 166 hours of human-recorded trajectories spanning three operating systems. By recording trajectories for closely related instructions, we can construct negative tasks by swapping the instructions. This paired design evaluates judges on their ability to distinguish a successful trajectory from one that completed a similar (but incompatible) request. We release the benchmark under three splits: a frontier split, AgentHorizon (AH), a simplified split, AgentHorizon-Simple (AH-S), and a development split, AgentHorizon-Development (AH-D). We further evaluate eleven judges by (1) directly passing the full trajectory (with up to 300 screenshots and actions), and (2) using them as coding agents across five agent harnesses. We find that our best agentic judge, GPT-5.5, achieves 80.9% balanced accuracy on the AH subset. We find that tool-use improves certain models but results in worse performance for open-weight models, and that judges differ drastically in their ability to accept a valid trajectory and reject failed ones. Our findings highlight the need for judges that are capable of locating and verifying often hidden evidence that a task was properly completed inside long interaction histories.

Xing Han Lù, Dheeraj Vattikonda, Sina Hajimiri +7

ServiceNow Research · McGill University · Mila – Quebec AI Institute +5

Computer-Use Agent Benchmarks Long-Horizon Agent Evaluation LLM-as-a-Judge

Oct 7, 2026cs.CL

Judging in Latent Space: Efficient Generative Reward Modeling via Semantics-Preserving Compression

Reward modeling often requires jointly representing and reasoning over multiple evaluation criteria, yet verbalizing this process token by token can incur substantial inference cost. Recent work on latent reasoning suggests that continuous states may support this computation more compactly. We introduce LatentGRM, a latent evaluation framework built on semantic chunking, compression, and reconstruction. By using the structure of rubric-guided evaluations to guide compression, LatentGRM learns compact continuous trajectories that support autonomous pairwise judgments without generating textual assessments. A separate interpreter reconstructs evaluation text from these trajectories, providing an offline view of the information retained under compression. Under matched training data and backbones, LatentGRM achieves competitive aggregate preference accuracy relative to explicit Supervised Fine-Tuning (SFT) judges at both 4B and 8B scales. Across four benchmark domains, LatentGRM-8B compresses evaluation trajectories by 8.9--9.2x and reduces total judge inference time by 6.1--7.0x at vote@5. Controlled rubric interventions show that criterion-dependent preference information is carried through the latent sequence. Together, these results demonstrate that continuous latent evaluation can substantially reduce inference cost while preserving competitive judgment quality.

Mingqing Yuan, Xiaobo Liang, Junwei Yang +5

Soochow University · University of Cambridge · Chalmers University of Technology +3

Reward Modeling LLM-as-a-Judge Continuous Latent Reasoning Efficient Language Model Inference

Oct 7, 2026cs.AI

A Tale of Two Error Categories: Exploring Concealed Trade-Offs in the Errors of Automated Judges in Evaluation of Uncertainty Quantifiers

The wide adoption of LLMs across broad NLG applications heightens the importance of providing users with the means to avert errors and hallucinations. Uncertainty quantification is poised to fill that gap; with low uncertainty (high confidence), as a proxy for correctness, allowing users to be selective (e.g., reject low-confidence, likely incorrect responses). Correlation between confidence and correctness then serves as a useful criterion for evaluation of uncertainty quantifiers (UQs). But in NLG, where diverse responses can be adequate to a prompt, obtaining reliable correctness judgements is not simple, especially without human intervention. Errors in automated judgement are hardly avoidable and known to diminish the reliability of evaluation protocols (Santilli et al., 2025; Ielanskyi et al., 2025). In a meta-analysis of published work, we show that automated judgement is the present norm. Besides, automated judgements are rarely validated against human ones, and the validation of the UQ evaluation they automate is even rarer. With experiments in question answering, using 4 LLMs, human and automated judgements and 7 popular UQs, we find that i) a judge's performance can only coarsely predict the observed impact of its errors on the reliability of UQ evaluation, and that ii) judgement errors tend to misrepresent informative UQs most. We link these observations to patterns of correlation between confidence and categories of judgement error.

Evgenia Ilia, Wilker Aziz

University of Amsterdam

Uncertainty Quantification Automated Evaluation LLM-as-a-Judge

Oct 7, 2026cs.AI

Reliability of LLM Judges for Evaluating Entity Alignment

Entity Alignment (EA) identifies equivalent entities across knowledge graphs and is critical for knowledge base integration and ontology merging. Evaluating EA systems at scale requires expensive expert annotation, making systematic assessment across diverse domains practically infeasible. LLM-as-judge evaluation offers a potentially scalable alternative, yet its reliability for structured prediction tasks like EA remains unstudied. We present the first systematic benchmarking study across three frontier models, three datasets, and four EA systems, using perturbation bias diagnostics, meta-evaluation across all dataset-judge-prompt combinations, and counterfactual label-flip tests. We identify anchor bias, a failure mode in which judges invert discrimination when the system's decision label is visible. Label exposure causally collapses judge discrimination (J-ROC-AUC 0.12-0.87), while a label-free protocol recovers near-ceiling capability on distinctive-name datasets (0.93-1.00) and significant recovery on biomedical pairs (0.93-0.95). Counterfactual experiments confirm causality (FSR 53-99%) and reveal a frontier model paradox: stronger judges exhibit greater label sensitivity, not less. A blinded two-annotator human evaluation (102 pairs, Cohen's kappa=0.902) confirms this mechanism directly. We release the first biomedical EA benchmark (MeSH-SNOMED CT, 15K pairs) and a reproducible auditing framework for LLM judge reliability in EA. Code and data are available at https://github.com/vaibhavalakshmiravideshik/llm-as-a-judge-entity-alignment.

Vaibhava Lakshmi Ravideshik, Mayank Kejriwal

University of Michigan, Ann Arbor · Information Sciences Institute, University of Southern California

LLM-as-a-Judge Bias in LLM-as-a-Judge LLM Reliability Knowledge Graphs

Oct 6, 2026cs.LG

Conditional Accuracy Profiles: Diagnosing LLM Judges across Deployment Conditions

LLM-as-judge is now a standard tool for scalable evaluation, but judge performance is still often summarized by a single accuracy number. This aggregate view hides the deployment conditions under which a judge succeeds or fails. We introduce \textbf{Conditional Accuracy Profiling} (CAP), a post-hoc diagnostic framework that decomposes pairwise LLM-judge accuracy into eight conditions organized into content sensitivity, robustness, and rationale quality. CAP is benchmark-agnostic: it can be applied directly when a benchmark provides the required annotations, approximately through task-subset proxies, or through controlled augmentation when perturbation pairs can be generated. We instantiate CAP on seven LLM judges across six pairwise judging benchmarks, including \textsc{judgerEva-Standard}, a controlled testbed we created to support all eight conditions. CAP exposes profile differences hidden by aggregate accuracy: on \textsc{judgerEva}'s judge-independent Hard-Constructed subset, the two judges most sensitive to omitted qualifications rank in the bottom three of seven by overall accuracy, so omission sensitivity is not predicted by aggregate accuracy. Across benchmarks, Position Robustness shows the strongest rank stability (mean Spearman

\barρ{=}0.87

) but is itself fragile under JudgeBench-Pro adversarial stress, showing the largest mean accuracy drop among the shared conditions, though the dominant degradation channel varies by judge. Condition-level profiles provide a more actionable basis than aggregate accuracy for selecting LLM judges.

Wenqi Li, Bin Liu, Mindi Ruan +3

University at Albany, State University of New York · West Virginia University

LLM Evaluation LLM-as-a-Judge Language Model Robustness

Oct 6, 2026cs.AI

AdaGuard: Enhancing Safety and Policy Compliance with Reasoning-Enabled LLM-As-A-Judge Guardrails

Enterprise generative AI applications require robust safety mechanisms that can accommodate diverse risk postures, evolving policies, and varying latency constraints. Current guardrail solutions often suffer from rigidity, relying on fixed policy sets and offering limited transparency or reasoning flexibility. We present Adaguard, an adaptive LLM-as-a-Judge framework designed to address these challenges through dynamic policy enforcement and adaptive reasoning-budget allocation. Built using supervised fine-tuning (SFT) and reinforcement learning (GRPO), AdaGuard generalizes to user-defined safety and compliance policies at runtime without requiring frequent model updates. A core innovation of our approach is the ability to dynamically infer the complexity of input-policy pairs, allowing the model to switch between high-speed black-box inference and explainable, reasoning-enabled moderation. This flexibility enables developers to balance stringent latency requirements with the need for actionable transparency. This adaptive capability allows AdaGuard to rival other guardrail and frontier models several times its size, while its auto-reasoning mode recovers the accuracy of always-on reasoning at a fraction of the latency

Melissa Kazemi Rad, Sihui Dai, Isha Slavin +6

Capital One

LLM Guardrails LLM Safety Content Moderation LLM-as-a-Judge

Oct 6, 2026cs.CL

Same-Number Citation Swaps: Stress-Testing Jev as a Financial Evidence Judge

Financial reports repeat values across periods, metrics and accounting lines, allowing an LLM-generated calculation to be numerically correct while citing the wrong financial role. We evaluate what probabilistic evidence verification adds beyond number matching using Jev as a source-support verifier for GPT-4.1-mini calculation traces. A signed-number-at-pointer baseline explains most recovery over exact quotation checks. To isolate the remaining role-recognition problem, we hold operands and arithmetic fixed, move citations between same-number cells, and retain controls that express equivalent facts. These contrasts reveal both wrong-role citations that pass and valid alternative citations that are withheld. Explicit column labels improve selected wrong-role decisions while also lowering support for some equivalent evidence. A constructed follow-up on 36 new source pages, labeled by a non-author reviewer, extends this evaluation and exposes the same tradeoff between detecting role errors and retaining valid citations. The contribution is a controlled evaluation that identifies what a probabilistic financial verifier distinguishes when numerical matching is held fixed. For LLM-based financial assistants, it makes numerical correctness, cited-role support and acceptance outcomes separately assessable.

Chuhong Xu, Bo Su, Ziyao Chen +3

Sofia University · Indiana University · University of California, San Diego +2

LLM-as-a-Judge Citation Verification

Oct 6, 2026cs.CL

The Labeling Problem in Hallucination Detection Benchmarks: An Empirical Evaluation

In recent years, several methods for detecting when large language models (LLMs) hallucinate have been developed. These methods are often benchmarked with open-domain question answering (QA) datasets containing questions and corresponding short reference answers. First, an LLM is used to generate answers to questions within the QA dataset. Then, some automated labeling strategy is used to label these answers as hallucinated or not by comparing them with the reference answers in the dataset. This evaluation setting creates a methodological ambiguity between two criteria: reference faithfulness (whether the answer is fully supported by the reference) and factual correctness (whether the answer is free from contradictions and factually false specific claims). In practice, automated labelers may apply the former criterion even when the intended target is the latter. We study this potential criterion mismatch using 900 human-labeled question-answer pairs spanning three commonly used QA datasets and three generator models, with labels targeting answer-level factual correctness. We evaluate lexical similarity metrics, a reference-entailment NLI baseline, and seven LLM judges under controlled prompt variants as automated labelers. Our experiments reveal substantial disagreement both among automated labeling strategies and between these labels and human annotations. Many strategies also exhibit strong directional error biases, and for most judge-generator pairs, replacing a faithfulness-oriented prompt with a factual-correctness prompt improves agreement with human annotations and reduces false-positive dominance, indicating that automated hallucination labels depend strongly on how the target criterion is specified. Label-source choice should therefore be considered a fundamental part of benchmark design and made explicit, validated, and matched with the benchmark goal.

Jorma Valjakka, Juhani Kivimäki, Juha Mylläri +1

Department of Computer Science University of Helsinki

LLM Evaluation LLM-as-a-Judge LLM Hallucination Detection

Oct 6, 2026stat.ML

Trustworthy Method Comparison with AI Judges: Estimation and Design under Order, Batch, and Aggregation Effects

Large language models (LLMs) are increasingly used as judges for automated AI evaluation. A common practice is to randomize prompt sequences and average the resulting scores, but its statistical validity remains unclear. We show that LLM evaluation mechanisms can be approximated by a class of Markov generalized linear mixed models (GLMMs), supported by out-of-sample predictions across three major commercial LLMs. Using a first-order Markov GLMM, we study leaderboard ranking and group comparison. For leaderboard ranking, randomize-and-average selection is consistent under a mild separation condition, and a Williams square design can improve efficiency when item qualities are close. For group comparison, naive averaging can yield inconsistent conclusions about differences in group-level quality because of the response model's nonlinearity. Empirical results further support the validity of the proposed model-based inference beyond the first-order theory, including settings with higher-order sequence memory. We illustrate the approach in an application where AI judges compare two graphical model estimation methods.

Tianxi Li, Jie Ding

School of Statistics, University of Minnesota

LLM Evaluation LLM-as-a-Judge Mixed-Effects Models

Oct 5, 2026cs.LG

CLM-as-a-Judge: Evaluating an Open Contrastive Decision Model on Public Judge Benchmarks

An open contrastive decision model is near chance as a judge on the hard public benchmarks: Contrastive-LM/CLM-v0.1-8B scores between 0.351 (best- of-four, chance 0.250) and 0.593 (pairwise, chance 0.500), is statistically indistinguishable from coin flipping on RM-Bench and JudgeBench, and answers every HaluEval item with one constant label, matching the trivial always-first baseline at 0.581. Judges with the same parameter count score far higher everywhere: a reward model reaches 0.764 to 0.976 and a generative judge 0.611 to 0.778, and every gap to CLM is significant after Benjamini-Hochberg correction. Two properties do work. Raw confidences are overconfident by up to +0.401, yet one pooled temperature fit on held-out calibration items repairs expected calibration error to at most 0.062, and the repaired confidence ranks the model's own errors above chance on three of six benchmarks. The decision order-flip rate is 0.0002 against 0.2188 for the generative judge, and the length-preference shift is -0.023 against -0.217. The confidence-gated cascade, however, escalates between 0.923 and 1.000 of items to the strong judge at the preregistered 0.97 retention bar: calibrated confidence about a near-chance judge has almost nothing to keep. The design: five public preference benchmarks and one hallucination benchmark with real labels, scored under a preregistration frozen before any test item was seen, against generative, reward-model, and trivial baselines, with per-item predictions released.

Gowthamkumar Nandakishore

Independent researcher

Contrastive Learning LLM Evaluation LLM-as-a-Judge Confidence Estimation in Language Models LLM Reliability

Oct 5, 2026cs.LG

Will the Judge Flip? Predicting Position-Sensitive LLM Judgments from Residual Stream Activations

The order in which candidate responses are presented can change an LLM judge's verdict. Detecting such a position flip ordinarily requires judging each pair in both orders, which doubles the number of judgments. We investigate whether residual stream activations recorded immediately before the initial verdict can predict a flip. We use nested grouped cross-validation to evaluate regularized linear probes on 534 JudgeBench pairs for three Qwen3 judges and Llama-3.1-8B. The linear probes achieve AUROCs of .621-.850 and outperform a combined baseline that uses verbalized confidence, verdict-label logits, response lengths, and the judge's initial choice by .062-.113 AUROC. Linear probes trained on JudgeBench and then frozen achieve AUROCs of .685-.853 on 1,802 MT-Bench comparisons without MT-Bench fitting or recalibration. These results show that pre-verdict activations support prediction of susceptibility to candidate order and outperform the non-activation predictors evaluated here.

Hashmath Shaik, Gnaneswar Villuri, Alex Doboli

Department of Electrical and Computer Engineering Stony Brook University Stony Brook, NY 11794, USA

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

Oct 5, 2026cs.CL

JudgeMoE: Distributional Aggregation for LLM-as-a-Judge

When an LLM judge scores an output, its score distribution retains uncertainty and disagreement information that is lost after scalar compression. We introduce JudgeMoE, a lightweight aggregator that assigns example-specific weights to cached judge score distributions and fuses them before computing a final score. A protocol study shows that score-range choice is unstable across judge--dataset settings and that soft scoring usually outperforms hard decoding. On the original 10-cell benchmark, JudgeMoE improves mean Spearman over uniform log pooling by

+0.079

. Applying the same configuration to six additional cells yields a

+0.0393

mean gain over the strongest local single judge across 16 cells, with positive differences in 12/16 cells and a one-sided Wilcoxon signed-rank

p=0.0091

. Validation-based analyses further show that the preferred aggregation method depends on the task and judge pool.

Yiqi Liu, Joseph James, Yang Wang +3

The University of Manchester · The University of Sheffield · University of Pittsburgh +1

LLM Evaluation LLM-as-a-Judge Ensemble Learning Language Model Ensembles

Oct 4, 2026cs.LG

VERA: Verdict-Conditioned Reliability for Adaptive LLM Judges

Accurately estimating judgment reliability is a central challenge in adapting LLM judges to newly verified feedback while preserving previously learned behavior. However, existing approaches often rely on output-level confidence, which can be overconfident and poorly aligned with judgment correctness. We propose VERA, a VErdict-conditioned Reliability Axis that estimates reliability from hidden activations by distinguishing correct from incorrect judgments within each predicted-verdict group. Using VERA as a control signal, we develop a VERA-guided periodic adaptation framework that integrates reliability-ranked corrective updates, reliability-residual replay, and periodic refresh of the reliability directions. After VERA-guided adaptation on Chatbot Arena, 8B- and 14B-parameter judges outperform the strongest baseline on each of four held-out public benchmarks, with relative gains of up to 23.01%. The framework also improves focal-class recall by up to 16.1% relative to the strongest adaptive baselines on a separate proprietary temporal auditing task.

Qiushui Xu, Syamil Mohd Razak, Tao Yuan +1

Penn State University · Amazon

LLM-as-a-Judge LLM Fine-Tuning LLM Reliability

Oct 4, 2026cs.CL

How Much Do LLM-as-a-Judge Design Choices Matter? A Systematic Comparison of Prompt Designs, Rating Scales, and Models

Researchers increasingly use Large Language Models as judges (LLM-as-a-judge) to evaluate model outputs. Yet there are no standards for how to design these judges. Typically, researchers choose the prompt, rating scale, and model intuitively. If these choices change the judge's verdicts, two studies can reach different conclusions about the same facts. To address this risk and to provide an empirical basis for judge designs, we evaluate 10 reasoning models across multiple designs on two tasks: a scalar rating of sentence sentiment and toxicity (over 500 items per category), as well as a binary accuracy classification of question-answer pairs (n=600). For the rating tasks, despite judges showing significant disagreements with the human ground truth, the practical size of differences is small enough to consider most judges reliable (mean absolute deviation of 0.11 points on a 1 - 7 scale); toxicity judges even outperform standard classifiers. Judges are also highly accurate on average (96.5%) for the accuracy classification task. However, design choices can produce shifts: changing the rating scale alone can shift measured bias by up to 0.93 points (rating task), and while accuracy levels are rarely impacted, design choices consistently impact judge leniency (classification task; leniency drop of 28.9 percentage points when using detailed prompts, and up to 56.1 percentage points when switching models). Counterintuitively, lower reasoning effort affects neither accuracy nor leniency. Across both tasks, model identity is the dominant source of variance. These findings suggest that while LLM judges are broadly trustworthy in aggregate, design choices can be meaningful sources of variance. Given the growing reliance on automated evaluation in LLM research, we intend this study as a methodological reference for designing more robust and replicable LLM-as-a-judge pipelines.

Laurène Vaugrante, Thilo Hagendorff

AI Safety Research Group, University of Stuttgart

LLM Evaluation LLM-as-a-Judge LLM Reliability

Oct 1, 2026cs.CR

Evaluating and Improving the Robustness of Large Language Models to Input Sequence Variations

Large language models (LLMs) in production systems face prompt injections, trojans (backdoors), and manipulation of automatic quality metrics. This thesis develops models, methods, and algorithms for evaluating and improving LLM robustness to adversarial input sequence variations. We propose R_stab(f), a generative robustness metric based on the Jensen-Shannon divergence between per-step output distributions under small input perturbations. For localized attacks we prove V(h) <= 1 - R_class(h), where R_class(h) is the probability that a decision operator h keeps its decision under small perturbations. For non-localized attacks we propose a calibrated empirical model. For LLM-as-a-Judge systems we develop ASA, an adaptive evolutionary black-box attack that reaches an attack success rate (ASR) of up to 73.8%, with transfer between open models up to 62.6%. On Trojan Detection Challenge 2023 data (Pythia-1.4B), surrogate triggers reach REASR ~0.99 while recall of the true triggers is ~0.17 against a baseline of ~0.14. On SaTML CTF 2024 we systematize four classes of bypasses of multi-layer defenses, which reduce the ASR from 90% to 15-25%. Committees of 5-7 heterogeneous models reduce the ASR for Gemma-3-4B by 47-55 percentage points, to 19.3% with 7 models. For agentic systems based on the Model Context Protocol (MCP), we propose AttestMCP, which attests tool calls with HMAC-protected packets at under 0.1 ms per call, and the Commit Boundary isolation pattern. On the MCPBench benchmark of 847 scenarios they reduce the average ASR from 53.7% to 12.4%. The methods are implemented in the JudgeGuard and TrojanArmor software suites and the MCPSec module.

Narek Maloyan

Language Model Robustness Adversarial Attacks on LLMs LLM Agent Security Backdoor Defense in LLMs LLM-as-a-Judge

Oct 1, 2026cs.CL

Old Ideas, Novel Problems: The Instability of LLM-Based Novelty Evaluation

Automated ideation systems are often evaluated on the novelty of the ideas they produce, and that judgment is increasingly delegated to large language models. Such judges are typically built ad hoc and validated, if at all, on human-authored papers rather than on the generated ideas they are meant to score. So, how do novelty judges perform? Not well. We present a systematic controlled study of novelty evaluation design choices. We first build an evaluation set automatically, mining OpenReview for passages where reviewers explicitly affirm or dispute a paper's originality and keeping only submissions with unanimous agreement at the extremes of their research area; we pair these with ideas from a vanilla LLM generator. Across six judges, we find that small prompt design choices have large consequences; e.g., simply telling the judge that reviewers found one idea novel and the other not can change its verdict on more than half of the identical idea pairs it is shown, shifting pairwise accuracy by over 50 points and occasionally pushing it below chance. The same change helps one judge and hurts another. Retrieval and larger reasoning budgets help little, and two purpose-built novelty evaluators are outperformed by our cheapest prompted baseline. These results raise questions about reported novelty gains of automated ideation systems, and call for robust novelty evaluation methods.

Noy Sternlicht, Simra Shahid, Peter Jansen +3

Hebrew University of Jerusalem · Allen Institute for AI · Microsoft +2

Prompt Sensitivity LLM-as-a-Judge Automated Evaluation

Oct 1, 2026cs.CV

Do MLLM Judges Judge the Edit? Auditing Bias in Image Editing Evaluation with Verified Quality Preservation

Multimodal large language models (MLLMs) are increasingly used as automated judges for instruction-based image editing and as reward signals for model training. However, systematically auditing whether these judges are influenced by cues irrelevant to editing quality is challenging because visual interventions may themselves alter the quality being evaluated. A judgment shift can therefore be attributed to bias only when the intervention is verified to preserve the underlying editing quality. To address this challenge, we introduce EditJudgeBias, a counterfactual benchmark with verified quality preservation, comprising 1,196 real editing samples and 13 cues injected across four evaluation sites. We verify quality preservation for the requested edit using calibrated multimodal validators, controls, and human inspection. We then audit five MLLM judges along three complementary dimensions: invariance to quality-preserving cues, agreement with human judgments, and stability of pairwise preferences. Importantly, observed shifts are evaluated against each judge's own zero-dose and re-query noise floors rather than against zero. Experiments show that quality-preserving cues move every judge beyond its own noise. Fabricated majority opinions increase ratings, irrelevant visual elements cause larger shifts than whole-image manipulations, and swapping candidate order reverses up to 60.9% of pairwise decisions. Edit-region cues also tend to reduce human agreement. The three measures characterize judges differently, showing that robustness cannot be captured by a single metric.

Yuan Huang, Zirui Song, Xiuying Chen

Northeastern University · Mohamed bin Zayed University of Artificial Intelligence

VLM Robustness LLM-as-a-Judge Image Editing Evaluation Counterfactual Evaluation of VLMs LLM Auditing Bias in LLM-as-a-Judge+1

Oct 1, 2026cs.LG

Range-GRPO: Policy Optimization via Pairwise Relations among Reward Intervals

As the use of large language models (LLMs) expands, post-training has become increasingly important for adapting them to downstream tasks. However, obtaining reliable supervision remains costly, especially in domains without reference answers or executable verifiers. LLM-as-a-Judge provides scalable pseudo-rewards for unlabeled responses, but a single point score does not explicitly represent reward uncertainty. This motivates representing pseudo-rewards as conformally calibrated reward ranges. We propose Range-GRPO, a semi-supervised post-training framework that combines limited labeled data with unlabeled prompts. In Group Relative Policy Optimization (GRPO), learning signals depend on relative reward comparisons within each rollout group. The proposed objective compares reward ranges pairwise rather than reducing them to point rewards, allowing interval uncertainty to affect both the magnitude and direction of these signals. Our theoretical analysis characterizes this distinction and shows that the proposed objective recovers the Dr.GRPO advantage when all reward ranges collapse to points. Empirically, Range-GRPO achieves the highest in-distribution and out-of-distribution average performance among the evaluated semi-supervised methods while requiring fewer training resources.

Ryunyi Lee, Kangjun Noh, Somin Kim +2

Yonsei University

Reward Modeling LLM-as-a-Judge Group Relative Policy Optimization RL Post-Training Semi-Supervised Learning

Oct 1, 2026cs.CV

When the Judge Acts: Auditing VLM-Guided Image Selection on Culturally Situated Prompts

Vision-language models (VLMs) increasingly act as judges that pick the best of several generated images, so their choices decide what users see. Such judges are usually validated by score agreement with human ratings, not by the images they return. We audit VLM judges as decision-makers: on 300 culturally situated prompts, we compare the returned image with human ratings the judge never sees and with random choice from the same candidates, and repeat every decision with the candidates reordered. A 4B-parameter judge barely beats random and falls short of a CLIP similarity baseline. It picks the first image shown in 49% of calls (chance: 28%), and reordering changes its choice on 60% of prompts. For this judge, agreement across orders is informative: decisions that survive reordering are much better than random, whereas agreement with a weaker second judge keeps the wrong ones. An 8B judge shows almost no position bias and outperforms CLIP, yet for it the same filter mostly discards good decisions. Agreement helps only when it targets the judge's failure mode, so filters must be re-audited whenever the judge changes. The 4B judge's slight rise in stereotype ratings is no longer detectable after aggregating across orders or with the larger judge.

Huichan Seo

Independent Researcher

VLM Evaluation LLM-as-a-Judge Vision-Language Model-as-a-Judge

Oct 1, 2026cs.CL

AGO AI Quality Gate: Evidence-First Release Decisions for Retrieval-Augmented Generation

Enterprises adopting retrieval-augmented generation (RAG) face a recurring operational decision: promote, revise, or block a system version. The evidence is incomplete and the metrics come from fallible LLM judges. We report on AGO AI Quality Gate (AGO), an evidence-first quality-gate framework deployed in industrial RAG assessment engagements. AGO integrates four key components: a four-state decision model that treats missing data and judge errors as explicit outcomes; layered scoring combining deterministic checks, local guardrails, and structured LLM evaluation; a stratified beta-binomial gate that quantifies regression risk probabilistically; and a mandatory meta-evaluation protocol to validate the LLM judge before it influences decisions. Since engagement data is proprietary, we evaluate the judge layer on RAGBench, a public benchmark of 100k annotated RAG traces across 12 datasets. On identical stratified test samples (N=1200 per judge), a low-cost judge (gpt-4.1-nano) detects non-adherent answers barely above chance (AUROC 0.603 [0.570, 0.634]), despite producing flawless protocol output, while gpt-4o reaches 0.783 [0.756, 0.807] -- yet its per-domain performance still ranges from 0.62 to 0.88. A fixed-seed gate study spanning regression, no change, and improvement quantifies unsafe promotion, false-alarm cost, and improvement throughput. Under regression, the decision-grade profile reduces unsafe promotion to 22.2%-35.1%, against 29.3%-41.8% for a naive gate. These results support the design choices that judge quality must be measured per engagement and that point estimates alone are not a release decision.

Giulio Zeloni, Enrico Lo Conte, Salvatore Rionero +3

Protom Group S.p.A., Napoli, Italy

LLM-as-a-Judge RAG Evaluation

Sep 30, 2026cs.AI

Better Deck or Different Judge? Evaluating Agentic Harness Gains in Corporate and Investment Banking

Corporate and investment banking teams use presentations to support credit decisions and advise clients on financing and transactions. Producing these decks requires reconciling financial data, tracing sources and turning analysis into a recommendation. We retrospectively study the development of an agentic harness combining a 27B language model, financial calculations, narrative templates and validation checks. LLM judges guide engineering changes and assess the resulting decks, raising the question of whether higher scores reflect better documents or changes in grading. In shared-session text-only grading with template markers removed, five judges score the complete system 20.4 to 33.6 points out of 95 above the same model generating directly from a short prompt. Every judge scores the system higher on all seventeen development deliverables. Margins against direct Opus generation from a short prompt range from -4.7 to +0.8 points. Judges agree on broad progress across development rounds but agree less on final-deck rankings than on pooled scores. Repeated grading also shifts scores on unchanged decks, making small improvements difficult to distinguish from judge variability.

Ludovic Gibert, Matis Despujols, Andre-Louis Rochet

AIData2Action · TW3 Partners

LLM-as-a-Judge Long-Form Document Generation LLM Agent Evaluation LLM Agent Harnesses

Sep 30, 2026cs.CL

LLM-as-a-Judge for Low-Resource Languages: Adapting Ragas and Comparative Ranking for Romanian

Evaluating Retrieval-Augmented Generation (RAG) systems remains a challenge for Low-Resource Languages (LRLs), where standard reference-based metrics fall short. This paper investigates the viability of the "LLM-as-a-Judge" paradigm for Romanian by adapting the Ragas framework using next-generation models (Gemini 2.5 and Gemini 3). We introduce AdminRo-Eval, a curated dataset of Romanian administrative documents annotated by native speakers, to serve as a ground truth for benchmarking automated evaluators. We compare three evaluation methodologies - direct scoring, comparative ranking, and granular decomposition - across metrics for Faithfulness, Answer Relevance, and Context Relevance. Our findings reveal that evaluation strategies must be metric-specific: granular decomposition achieves the highest human alignment for Faithfulness (96% with Gemini 2.5 Pro), while comparative ranking outperforms in Answer Relevance (90%). Furthermore, we demonstrate that while lightweight models struggle with complex reasoning in LRLs, the Gemini 2.5 Pro architecture establishes a robust, transferable baseline for automated Romanian RAG evaluation.

Claudiu Creanga, Liviu P. Dinu

Interdisciplinary School of Doctoral Studies, University of Bucharest, Romania · HLT Research Center, University of Bucharest, Romania · Faculty of Mathematics and Computer Science, University of Bucharest, Romania

LLM Evaluation LLM-as-a-Judge Language Model Generation Evaluation RAG Evaluation Automated Evaluation Low-Resource Language Processing

Sep 30, 2026cs.CL

RAIM: Robust Aggregation of Inexpensive Models for Hallucination Detection

Automatic evaluation of faithfulness increasingly relies on a large language model acting as a judge, yet the most reliable judges are proprietary frontier models, costly and ill-suited to high-throughput monitoring. We investigate whether a panel of cheap open-weight judges (4--9B) can be aggregated to stand in for a frontier one, what the substitution sacrifices, and when it is worth making. We propose RAIM, an aggregation scheme robust to the members' correlated errors, coupling a cross-fitted stacked logistic regression with an admissibility test that, read from the members' own outputs, identifies when aggregating them improves on their best member and stays within reach of the frontier judge. We instantiate RAIM with ten judges from disjoint families across eight faithfulness benchmarks. Against Claude Sonnet, the panel retains a median 93% of its Cohen's

κ

and gives up only 2.9 points of balanced accuracy on average; read as paired differences, it clearly improves on one benchmark and clearly worsens on three (only two by a non-negligible margin), leaving four unresolved. At a sixty-fourth of the frontier's inference price, the operative expense is a one-time in-domain calibration on 50--100 labelled records. The panel is also competitive with purpose-trained detectors on their home benchmarks (within 1.3 accuracy points of GPT-4o and 1.9 of the LLM-AggreFact leader), and beats the strongest one we reran by 6 points on our grounded sets. Whether aggregation pays depends on the members themselves: where several capable members err on different items, the panel improves on its best judge and approaches the frontier; where one dominates, the stacker recovers the leader, and only there does the frontier remain materially ahead. Both conditions are read off the calibration set at no further cost, so a cheap panel can stand in for a frontier one wherever this audit admits it.

Elia Onofri, Roberto Di Pietro

Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division, King Abdullah University of Science and Technology (KAUST) Thuwal 23955, Saudi Arabia

LLM Evaluation LLM-as-a-Judge LLM Hallucination Detection Language Model Ensembles

Sep 30, 2026cs.CL

Training LLM Judges from Language Feedback via Position-Selective Self-Distillation

We study training LLM judges from natural language feedback, especially for subjective tasks where the verdict depends strongly on which evaluation criteria the judge invokes and how it weighs them. The dominant approach, outcome-supervised RL (e.g., GRPO), credits every token in the rollout with a single scalar determined only by the accuracy of the final verdict, providing no separate credit at the criterion-choice tokens and ignoring the rich language feedback (e.g., preference rationales) that naturally accompanies preference labels. Self-Distillation (SD) is one natural way to use this language feedback: the same model, conditioned on this feedback, acts as a teacher providing dense, position-level supervision. However, not all positions carry equally useful signal. Using the per-position entropy shift between teacher and student, we identify two regimes: context sharpening, where the teacher concentrates probability on a particular feedback-aligned criterion expression, and context spreading, where the teacher distributes probability across multiple feedback-aligned alternatives. We interpret these patterns as follows: sharpening encourages memorization of a particular criterion expression, whereas spreading promotes semantic understanding by preserving these alternatives. Motivated by this asymmetry, we introduce position masking based on the entropy shift that retains the lower tail of the entropy-shift distribution. Experiments show that masking higher-entropy-shift positions improves out-of-distribution generalization over naive SD. The resulting self-distilled judges outperform judges trained with outcome-supervised RL by 2-9 percentage points on the evaluated subjective subcategories, while remaining competitive on objective ones.

Ilgee Hong, Changlong Yu, Zhenghao Xu +5

Georgia Institute of Technology · Amazon · UC San Diego

LLM-as-a-Judge Language Model Distillation Context Distillation

Sep 29, 2026cs.CL

Pair Difficulty Matters: Rethinking Pairwise LLM-as-a-Judge Evaluation and Consistency

Large Language Model judges are widely used to rank texts and text-generating systems through pairwise comparison, and their reliability is typically assessed via three proxies: position bias, transitivity, and pairwise agreement (self- or human-labeled). Because these proxies drive judge selection and benchmarking, a substantial literature reporting that judges perform poorly on them risks steering practitioners away from otherwise capable evaluators. We argue this assessment is misleading. Under the Bradley--Terry geometry underlying pairwise aggregation, each proxy is dominated by close-rank-gap pairs, where inconsistency is information-theoretically expected and individual verdicts contribute little to the aggregate ranking; far-gap pairs carry the ranking signal but barely move the proxies. We formalize this argument and validate it in a controlled simulation and on two human-rated corpora: the proxies correlate only weakly with ranking accuracy against gold, and their predictive component concentrates in the far-gap regime. Judges should therefore be assessed on rank-gap-conditional metrics, ideally against human rankings. Code at https://github.com/brunobrocai/PairDifficulty.

Bruno Brocai, Maria Becker

Heidelberg University German Department

LLM-as-a-Judge Pairwise Preference Evaluation LLM Reliability

Sep 29, 2026cs.CV

Evaluating the Evaluators: Diagnosing Large Multimodal Models for AI-Generated Image Assessment

With the rapid advancement of text-to-image (T2I) generation, robust evaluation becomes critical yet challenging, as traditional metrics fail to capture fine-grained alignment and generative artifacts. While large multimodal models (LMMs) are increasingly adopted as evaluators, existing benchmarks typically study semantic understanding, quality perception, and authenticity identification in isolation, while largely neglecting responsibility detection. This leaves a gap in unified and comprehensive validation. To bridge this gap, we introduce SQUARE-Bench, a comprehensive benchmark that systematically evaluates LMM capabilities as evaluators of AI-generated images across four aspects: Semantics, Quality, Authenticity, and Responsibility. SQUARE-Bench introduces a granular taxonomy of 38 sub-dimensions to evaluate nearly 10K AI-generated images sampled from 22 diverse models, ranging from legacy to state-of-the-art generators, complemented by over 3K real-world images. The images are annotated with curated question-answering pairs. Extensive experiments on 23 LMMs reveal that top proprietary models, such as Gemini-3-Pro, already outperform the individual human expert baseline. However, the performance gap between models remains significant, exhibiting notable disparities in fine-grained inference and domain-specific robustness. Beyond benchmarking, we conduct a proof-of-concept study of LMM-guided iterative editing, in which dimension-specific LMMs provide diagnostic feedback to fixed image editors. The resulting guided system yields selective improvements in semantics, authenticity, and responsibility, while exhibiting a consistent visual-quality trade-off. SQUARE-Bench can serve as both a diagnostic tool for characterizing LMM evaluator capabilities and studying their use in T2I generation refinement. The benchmark and dataset will be released upon publication.

Yu Zhao, Jiarui Wang, Huiyu Duan +5

Shanghai Jiao Tong University · Dalian University of Technology

LLM-as-a-Judge Image Generation Evaluation Multimodal Model Evaluation T2I Generation Evaluation