Bias and Fairness in Large Language Models: A Survey
Open accessThe DOI matches its Crossref record; 313 of the 313 references it cites were found in a registry or an index; 151 of the 183 sources its claims cite were checked for support.
Abstract
Rapid advancements of large language models (LLMs) have enabled the processing, understanding, and generation of human-like text, with increasing integration into systems that touch our social sphere. Despite this success, these models can learn, perpetuate, and amplify harmful social biases. In this article, we present a comprehensive survey of bias evaluation and mitigation techniques for LLMs. We first consolidate, formalize, and expand notions of social bias and fairness in natural language processing, defining distinct facets of harm and introducing several desiderata to operationalize fairness for LLMs. We then unify the literature by proposing three intuitive taxonomies, two for bias evaluation, namely, metrics and datasets, and one for mitigation. Our first taxonomy of metrics for bias evaluation disambiguates the relationship between metrics and evaluation datasets, and organizes metrics by the different levels at which they operate in a model: embeddings, probabilities, and generated text. Our second taxonomy of datasets for bias evaluation categorizes datasets by their structure as counterfactual inputs or prompts, and identifies the targeted harms and social groups; we also release a consolidation of publicly available datasets for improved access. Our third taxonomy of techniques for bias mitigation classifies methods by their intervention during pre-processing, in-training, intra-processing, and post-processing, with granular subcategories that elucidate research trends. Finally, we identify open problems and challenges for future work. Synthesizing a wide range of recent research, we aim to provide a clear guide of the existing literature that empowers researchers and practitioners to better understand and prevent the propagation of bias in LLMs.
Overview
CoChat Summary
TL;DR
This survey consolidates, formalizes, and taxonomizes bias evaluation and mitigation for large language models, covering metrics, datasets, and mitigation techniques. It proposes three taxonomies and identifies open problems around power imbalances, fairness conceptualization, evaluation standards, mitigation scalability, and theoretical guarantees.
Key findings
- The survey proposes three taxonomies: bias evaluation metrics organized by what they use from a model (embeddings, probabilities, generated text); bias evaluation datasets organized by structure (counterfactual inputs vs. prompts); and bias mitigation techniques organized by intervention stage (pre-processing, in-training, intra-processing, post-processing).
- It formalizes social bias for LLMs as representational harms (e.g., derogatory language, disparate system performance, erasure, exclusionary norms, misrepresentation, stereotyping, toxicity) and allocational harms (direct and indirect discrimination), and introduces fairness desiderata including fairness through unawareness, invariance, equal social group associations, equal neutral associations, and replicated distributions.
- It reports limitations across metric classes: embedding-based and probability-based metrics may be only weakly or inconsistently correlated with downstream task bias, and generated-text metrics can be sensitive to decoding parameters; it also highlights reliability and validity issues in widely used datasets.
- It consolidates publicly available datasets, with reported sizes including CrowS-Pairs (1,508), StereoSet (16,995), BBQ (58,492), RealToxicityPrompts (100,000), BOLD (23,679), HolisticBias (460,000), and Bias NLI (5,712,066).
Methods
- Design: a comprehensive survey and taxonomy-building effort rather than an empirical study, synthesizing prior work on bias evaluation and mitigation for LLMs.
- Scope: focuses on bias issues in LLMs for English, with additional languages for machine translation and multilingual models, and restricts inclusion to works that propose novel closed-form metrics, datasets, or mitigation techniques and specify applicability to social bias or fairness.
- Analysis: organizes metrics, datasets, and mitigation techniques into taxonomies, formalizes several metrics and mitigation objectives with unified notation, discusses limitations of each class, and outlines open problems and recommendations.
Limitations
- The paper states that it aligns with a technical solutionist perspective rather than a critical theoretical one, and that its taxonomies are organized by technical implementation details instead of downstream usage contexts or harms.
- The paper states that it focuses narrowly on model training and evaluation, rather than a more holistic pipeline including problem formulation, data collection, and deployment/integration.
- The paper states that it is limited to English language papers; evident from the text, it also does not report a systematic search protocol, search dates, or screening counts.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures










Verification
Claims the paper makes
121 Valid · 12 Not valid · 18 Not checked · 106 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Despite the growing emphasis on addressing these issues, bias and fairness research in LLMs often fails to precisely describe the harms of model behaviors: who is harmed, why the behavior is harmful, and how the harm reflects and reinforces social principles or hierarchies (Blodgett et al. 2020).”
What the sources say
- Partly supportedcites Blodgett et al. 2020 matched to “Language (Technology) is Power: A Critical Survey of “Bias” in NLP” (2020) sourceExperimental — accuracy not yet measured
Section 3.1 (Motivations; page not provided) finds that papers on bias in NLP often fail to explain which model behaviors are harmful, how, to whom, and why. Sections 4.1–4.2 call for attention to social hierarchies and explicit normative reasoning. The survey concerns NLP research broadly, however, and does not establish that this failure is common specifically in LLM research. (LLM, conf=0.86)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Similar to the shortcomings of embedding-based metrics, Delobelle et al. (2022) and Kaneko, Bollegala, and Okazaki (2022) point out that probability-based metrics may be only weakly correlated with biases that appear in downstream tasks”
- Also stated in: § 3.6 Recommendations · p. 27 · plain
“Probability-based metrics also show weak correlations with downstream biases.”
What the sources say
- Supportedcites Delobelle et al. 2022 matched to “Measuring Fairness with Biased Rulers: A Comparative Study on Bias Metrics for Pre-trained Language Models” (2022) source
Section 3 describes CrowS-Pairs as a pseudo-loglikelihood-based metric. Section 4.4 reports that CrowS-Pairs correlates less with two measures of bias in downstream tasks, supporting the hedged claim that probability-based metrics may correlate weakly with downstream bias. The finding does not establish that all probability-based metrics do so. (LLM, conf=0.90)
- Supportedcites Kaneko, Bollegala, and Okazaki 2022 matched to “Debiasing isn't enough! -- On the Effectiveness of Debiasing MLMs and their Social Biases in Downstream Tasks” (2022) source
The Introduction (p. 1) identifies intrinsic measures as using likelihood scores assigned by MLMs and reports only weak correlations with downstream, task-specific bias measures. Section 5.2 (p. 8) likewise finds weak and inconsistent correlations between intrinsic and extrinsic bias scores. (LLM, conf=0.93)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 3.6 Recommendations · p. 27 · plain
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Many approaches, for instance, assume some implicitly desirable criterion ( e.g. , a model output should be independent of any social group in the input), but do not explicitly acknowledge or state the normative social values that justify their framework.”
What the sources say
- Partly supportedcites Blodgett et al. 2021 matched to “Stereotyping Norwegian Salmon: An Inventory of Pitfalls in Fairness Benchmark Datasets” (2021) sourceExperimental — accuracy not yet measured
Section 2.3 describes assumptions about ideal model behavior, including equal probabilities for sentences about different target groups. Section 6 says the benchmarks make implicit decisions about stereotyping that are not explicitly discussed or justified. The paper examines four stereotyping benchmarks, however, rather than establishing this claim about many bias and fairness approaches generally or specifically about their normative social values. (LLM, conf=0.84)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Cabello, Jørgensen, and Søgaard (2023) illustrate that associations between the representations of protected attribute and other words can be independent of downstream performance disparities, if certain assumptions of social groups' language use are violated.”
What the sources say
- Supportedcites Cabello, Jørgensen, and Søgaard 2023 matched to “On the independence of association bias and empirical fairness in language models” (2023) source
The Bias section defines association bias as differences in how demographic-group terms are encoded and empirical fairness as performance differences across social groups. Section 5 says the two can be independent and explains that linking them relies on the In-Group Affinity Assumption about how groups use those terms; it calls that assumption problematic. No page numbers are provided in our copy. (LLM, conf=0.93)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Each stakeholder brings perspectives that may specify different fairness constraints for the same application and setting.”
What the sources say
- Partly supportedcites Jernite et al. (2022) matched to “Data Governance in the Age of Large-Scale Data-Driven Language Technology” (2022) sourceExperimental — accuracy not yet measured
Section 2.2 (p. 4) says participants bring different perspectives and warns that relying only on their values could exclude views more relevant to other language users. The paper discusses nondiscrimination and local expressions of values, but does not say that stakeholders specify different fairness constraints for the same application and setting. (LLM, conf=0.85)
- Not matchedcites Bender (2019)
An author–year citation the paper’s reference list doesn’t settle. We don’t search these by name.
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
71 more claims · 34 with their cited sources checked
Create a free account to see all 257 claims and every source verdict.
- Probably not a substantial claim. 151
- It may state two separate claims as one. 70
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“While this process mitigates some risks inherent to AI-generated content, it does not eliminate potential biases associated with proprietary large language models, which lack transparency regarding training data and internal parameters [10].”
From AI-Generated Multiple Mini Interview (MMI) Stations for Medical School Admissions: Psychometric Evaluation (2026) · cites this paper as [10]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Notably, there is growing attention on assessing LLM bias [63-64] and robustness [323-324].”
From A Survey of Large Language Models (2026) · cites this paper as [63]
The cited source says what the claim attributes to it.
In the citing paper
“these models pose serious risks, as the vast, unfiltered internet data used to train them can reinforce and amplify harmful biases, disproportionately harming marginalized communities (73).”
From Emergent social conventions and collective bias in LLM populations (2025) · cites this paper as [73]
The cited source says what the claim attributes to it.
In the citing paper
“Despite the use of various mitigation techniques, such as debiasing algorithms, curating balanced datasets, or incorporating fairness-focused training objectives, eliminating bias entirely is a persistent challenge141–143.”
From Current applications and challenges in large language models for patient care: a systematic review (2025) · cites this paper as [142]
The cited source says what the claim attributes to it.
In the citing paper
“Another major concern with LLMs is their propensity to amplify historical biases.”
From Evaluation of Large Language Models for Peer Review in Transplantation Research: Algorithm Validation Study (2026) · cites this paper as [30]
1 more claim about this paper
Create a free account to see every claim other papers make about this one.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.1 Counterfactual Inputs · p. 28
“| WinoBias+ | 1,367”
§ 4.1 Counterfactual Inputs · p. 29
“WinoBias+ (Vanmassenhove, Emmery, and Shterionov 2021) augments WinoBias with gender-neutral alternatives, similar to Winogender's neutral option, with 3,167 total instances.”
Citations
Cites
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325 · cited 6×
“, 2020), and GPT-4 (OpenAI, 2023); autoencoding models include BERT (Devlin et al., 2019), RoBERTa (Liu et al.”
BackgroundInfluential(2014) · Cited by 33,419
A Unified Approach to Interpreting Model Predictions
Lundberg, S., & Lee, S.-I. (2017) · arXiv · doi:10.48550/arxiv.1705.07874 · Cited by 23,707
Language Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372 · cited 2×
“, 2019), GPT-3 (Brown et al., 2020), and GPT-4 (OpenAI, 2023); autoencoding models include BERT (Devlin et al.”
Background"Why Should I Trust You?": Explaining the Predictions of Any Classifier
Ribeiro, M. T., Singh, S., & Guestrin, C. (2016) · Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2939672.2939778 · Cited by 17,555
RoBERTa: A Robustly Optimized BERT Pretraining Approach
Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., & Stoyanov, V. (2019) · arXiv · doi:10.48550/arxiv.1907.11692 · Cited by 16,515
Kaugeranna, R., Kaugeranna, E., & Aion, (. S. 4. 6. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.giscience.2023.43 · Cited by 13,690
Gauriat, C.-M., Pencolé, Y., Ribot, P., & Brouillet, G. (2024) · HAL (Le Centre pour la Communication Scientifique Directe) · doi:10.4230/oasics.dx.2024.27 · Cited by 12,586
Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., & Liu, P. J. (2019) · arXiv · doi:10.48550/arxiv.1910.10683 · Cited by 10,929 · cited 1×
“The rise and rapid advancement of large language models (LLMs) has fundamentally changed language technologies (e.g., Brown et al., 2020; Conneau et al., 2020; Devlin et al., 2019; Lewis et al., 2020; Liu et al., 2019; OpenAI, 2023; Radford et al., 2018; 2019; Raffel et al., 2020).”
BackgroundMeasuring individual differences in implicit cognition: the implicit association test
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998) · Journal of Personality and Social Psychology · doi:10.1037/0022-3514.74.6.1464 · Cited by 9,210 · cited 2×
“, family and occupation words), emulating the Implicit Association Test (Greenwald et al., 1998).”
Method
Research works printed but not linked: 5
- Barocas, 2019, Fairness and Machine Learning: Limitations and Opportunities
- Blodgett, Su Lin . 2021. Sociolinguistically Driven Approaches for Just Natural Language Processing. Ph.D. thesis. University of Massachusetts Amherst.
- Green, 2019, ”Good” isn’t good enough, 1
- Radford, 2019, Language models are unsupervised multitask learners, OpenAI Blog, 1, 9
- Wang, 2019, BERT has a mouth, and it must speak: BERT as a Markov random field language model, 30, 10.18653/v1/w19-2304 · doi:10.18653/v1/w19-2304
Cited by
Items 401–500 of 763
Benchmark Study for Evaluating the Korean Comprehension of Large Language Models
Seo, M., Jeong, Y., Lee, H., Im, H., & Jang, J. (2024) · The Journal of Korean Association of Computer Education · doi:10.32431/kace.2024.27.9.007 · Cited by 1
Prompts De-Biasing Augmentation to Mitigate Gender Stereotypes in Large Language Models
Chen, B., Binnewies, S., & Stantić, B. (2025) · Lecture notes in computer science · doi:10.1007/978-981-96-6008-7_19 · Cited by 1
“Navigli [18], Gallegos [5], had noted that large-scale pre-trained LLMs are typically trained on tens of billions of text sequences sourced from the internet, books, and media, which inherently contain biases and stereotypes related to gender, race, culture, and other societal attributes.”
InfluentialIdentifying Representation Bias in Large Language Models Used in Financial Sentiment Analysis
Sabuncuoğlu, A., & Maple, C. R. (2025) · IEEE Conference on Computational Intelligence for Financial Engineering & Economics · doi:10.1109/cifer64978.2025.10975731 · Cited by 3
“[9].”
Trustworthiness Evaluation of Large Language Models
Chen, P., & Liu, S. (2025) · Springer eBooks · doi:10.1007/978-3-031-76770-8_12 · Cited by 2
Measuring Political Bias in LLMs using Fine-Tuning RoBERTa Model
Rajathi, V., Singh, A. R., & Kumar, K. (2025) · International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS) · doi:10.1109/icssas66150.2025.11081114 · Cited by 1
“Muhammad Umar Nadeem [1] had proposed that detecting political bias in news headlines has been extensively researched using a variety of machine learning and natural language processing techniques.”
Jia, K., Qin, J., & Li, Y. (2025) · PLoS ONE · doi:10.1371/journal.pone.0330160 · Cited by 3
“Through transfer learning and fine-tuning, such models efficiently adapt to domain-specific classification tasks, significantly increasing recognition accuracy for feature-specific terms in complex unstructured texts [34].”
Design and Implementation of a Tactical Wargame Adversarial Platform Based on Lightweight SDK
Hu, R. H., Qi, W., Zhao, B. C., & Ma, H. (2025) · International Conference on Computer Information Science and Application Technology (CISAT) · doi:10.1109/cisat66811.2025.11181989 · Cited by 1
“Toolkits such as OpenAI Gym [7] , Petting Zoo [8] , and MAgent provide standardized environments for agent training in both single-agent and multi-agent settings.”
Brainstorming Brings Power to Large Language Models of Knowledge Reasoning
Qin, Z., Wang, C., Guo, J., Qin, H., & Jia, W. (2025) · IEEE International Conference on Multimedia and Expo (ICME) · doi:10.1109/icme59968.2025.11209475 · Cited by 4
“Despite their ability to handle various tasks, the reasoning biases resulting from a single model’s limited perspective remain a concern [3].”
Are Horses Always Strong and Donkeys Dumb? Animal Bias in Vision Language Models
Anas, M., Nadeem, M., Sohail, S. S., Cambria, E., & Hussain, A. (2025) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn64981.2025.11228584 · Cited by 1
“A comprehensive survey of bias in large language models can be found in [35].”
Developing an ethical framework for intelligence-AI
Bellaby, R. W. (2026) · AI and Ethics · doi:10.1007/s43681-026-01076-5 · Cited by 1
The Roles of AI in Tobacco and Nicotine Research and Regulation
Kalan, M. E., Dadipoor, S., Osibogun, O. I., & Taleb, Z. B. (2025) · Substance Use & Misuse · doi:10.1080/10826084.2025.2568154 · Cited by 2
“This can lead to cultural biases being rooted within the algorithms (Gallegos et al., 2024).”
Mccalla, D., Li, B., Jaeger, S., Jacques, J. G., Gonzalez, L., Jr., Silber, C. L., & Dykeman, C. (2026) · doi:10.2196/preprints.93498 · Cited by 1
AI adoption in Spain (2023–2025): A web-derived dataset based on LLMs
Pastor-Merino, A., Martínez-Barbero, X., & Domènech, J. (2026) · Data in Brief · doi:10.1016/j.dib.2026.112662 · Cited by 1
“However, methodological caution is required due to risks related to bias, prompt sensitivity, and hallucinations [ 12 , 13 , 10 ], underscoring the importance of prompt design [ 14 ] and transparent evaluation frameworks [ 15 ].”
Dycke, N., & Gurevych, I. (2026) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl.a.642 · Cited by 1
Hamila, S., Birchill, K., Cao, K., Hossain, M. N., Bullock, S., Hodgson, W., Harrison, J., & Leech, M. (2026) · JMIR Medical Education · doi:10.2196/86208
“While this process mitigates some risks inherent to AI-gener-ated content, it does not eliminate potential biases associated with proprietary large language models, which lack transparency regarding training data and internal parameters [10].”
Quantifying and Mitigating Self-Preference Bias of LLM Judges
Yang, J., Hu, Z., Qiu, C., Deng, Z., Jiao, X., & Zhou, T. (2026) · arXiv · doi:10.48550/arxiv.2604.22891
“The core assumption underlying this paradigm is fairness, namely the increasing model scale and stronger reasoning capabilities will naturally lead models to overcome bias and act as more objective evaluators[10, 11, 12].”
Trustworthy Medical Question Answering: An Evaluation-Centric Survey
Wang, Y., Wang, B., Mercer, R. E., Rudzicz, F., Roy, S. S., Ren, P., Chen, Z., & Wang, X. (2025) · arXiv · doi:10.48550/arxiv.2506.03659
Conformity in Large Language Models
Zhu, X., Zhang, C., Stafford, T., Collier, N., & Vlachos, A. (2024) · arXiv · doi:10.48550/arxiv.2410.12428
“Most biases studied in LLMs tend to be overt and domain-specific ( e . g ., gender, race, etc.) making them relatively easier to detect and mitigate (Gallegos et al., 2024; Ranaldi et al., 2024).”
Inspecting and Measuring Fairness of unlabeled Image Datasets
Görge, R., Möck, M., & Akila, M. (2024) · IEEE International Conference on Data Engineering Workshops (ICDEW) · doi:10.1109/icdew61823.2024.00031 · Cited by 1
“However, using embedding based metrics for bias evaluation of models as well as the use of reference data with high data quality are well known from text analysis [18].”
MethodDo AI Chatbots Provide an Outside View?Preprint
Shu, S., Das, S., Hernandez, D. P., Fayaz, O., Lei, J., Manavalan, P. K. M., Peng, X., Sakaguchi, N., & Suresh, S. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4874756 · Cited by 1
Towards Construction-Oriented Play for Vision-Diverse People
Rodriguez, A., Devasia, N., Pei, M., & Kientz, J. A. (2024) · International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS) · doi:10.1145/3663548.3675607
“However, while LLM’s ability to perpetuate culture through verbal feedback gives blind people access to private and on-demand feedback, we acknowledge the risks that come with using artifcial intelligence to provide opinions regarding culture [21].”
What does AI consider praiseworthy?
Peterson, A. (2025) · AI and Ethics · doi:10.1007/s43681-025-00682-z · Cited by 1
Unpacking the Outputs of Generative AI Platforms and Revealing Gender and Social Re-Presentations
Risi, E., Briziarelli, M., & Mohammad, Y. (2024) · Advances in human and social aspects of technology book series · doi:10.4018/979-8-3693-7235-7.ch012 · Cited by 2
Enhancing Annotated Bibliography Generation with LLM EnsemblesPreprint
Bermejo, S. (2025) · Qeios · doi:10.32388/wihg9q · Cited by 3
Aroeira: A Curated Corpus for the Portuguese Language with a Large Number of Tokens
Lira, T., Cação, F. N., de Souza, C. A., Valentini, J. P. H., Bollis, E., Oliveira, O., Almeida, R., Magalhães, M. L. D., Poloni, K. M., Oliveira, A., et al. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-79029-4_13
Opening a New Chapter for Computational Linguistics
Lu, W. (2025) · Computational Linguistics · doi:10.1162/coli_e_00552
MyRecommender: Personalized Activity Recommendation based on Negative Emotions Classification
Shafraz, Z., Ilangeshwaran, H., Shifan, M. R. M., Vithanage, C., Abeywardana, L., Rathnayake, S. C., & Ismath, R. (2024) · International Conference on Information Technology Research (ICITR) · doi:10.1109/icitr64794.2024.10857717
Narratives of Divide: The Polarizing Power of Large Language Models in a Turbulent WorldPreprint
Saqr, K. M. (2025) · Research Square · doi:10.21203/rs.3.rs-5950965/v1
Co-creating Intercultural Scenes With Gen AI Using a Case Study Framework
Hayward, P. G., Bunning, L., & Lages, W. S. (2025) · Advances in computational intelligence and robotics book series · doi:10.4018/979-8-3373-0502-8.ch007
Dube, S. S., Ndlovu, S., & Ncube, N. (2024) · Zimbabwe Conference of Information and Communication Technologies (ZCICT) · doi:10.1109/zcict63770.2024.10958402
Systematic Review of Fine-tuning Approaches for Large Language Models in the Context of Sinhala
Hansaka, S., & Sumanathilaka, T. G. D. K. (2025) · International Conference on Advanced Research in Computing (ICARC) · doi:10.1109/icarc64760.2025.10962880 · Cited by 2
“LLMs, despite their strengths, face limitations such as ”hallucination” and biases in their training corpus, which can lead to harmful results in low-resource languages like Sinhala [20], [21].”
Napiwotzki, I., Laue, J., Caldarone, F., Idahl, M., Hadler, U., Amrani, H., Hildt, E., Kahl, K. G., & Nejdl, W. (2025) · doi:10.2196/preprints.78138
Large Language Models for the Analysis of Project Proposals
Tsangko, I., Triantafyllopoulos, A., Kyriakidis, E., Margetis, G., & Schuller, B. W. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-93415-5_24
Judgment Prediction Models as an AI-driven Jury and its Ontology for E-courtPreprint
Chang, Y.-F. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5274505 · Cited by 1
Berlincioni, L., Cultrera, L., Becattini, F., Bertini, M., & Bimbo, A. D. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-92648-8_7 · Cited by 2
Navigating Trustworthiness in LLMs: An Examination of Privacy, Security, and Robustness
Dang, K., & Lai, P. (2025) · Lecture notes in computer science · doi:10.1007/978-981-96-6389-7_1
Systematic Review of Large Language Models and Future Directions
Solomon, E., Woubie, A., & Emiru, B. S. (2025) · International Conference on Information and Computer Technologies (ICICT) · doi:10.1109/icict64582.2025.00036
Wang, Y., Chen, B., & Sun, J. (2025) · Symmetry · doi:10.3390/sym17071076
Speculative AI Re-enactment of the Figurists' Encounters with the I Ching
Clarke, I. J., Masu, R., & Papatheodorou, T. (2025) · Proceedings of the ACM on Computer Graphics and Interactive Techniques · doi:10.1145/3736780
“Similarly, biases can be found in contemporary predictive systems, such as AI Large Language Models (LLM) [Gallegos et al. 2024].”
Embedding Conversational Safety in AI for Education
Buddarapu, V. N., & Lynch, C. F. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-98462-4_47
Barrit, S., Ranuzzi, G., Fetzer, S., Barajraji, M. A., Hadwe, S. E., Zanello, M., Ortler, M., O’Flaherty, J., Massager, N., Madsen, J. R., et al. (2025) · Acta Neurochirurgica · doi:10.1007/s00701-025-06610-8
“Though curation provides control over AI outputs, it cannot fully mitigate the inherent biases present in foundational LLMs [19, 34, 35].”
Fair Summarization in the Age of Large Language Models: A Comprehensive Survey
Mittal, S., Das, V., Yadav, V. H., & Singh, K. (2025) · International Conference on Intelligent Communication Technologies and Virtual Mobile Networks (ICICV) · doi:10.1109/icicv64824.2025.11086003
“Bias in LLMs [25] deviate the LLM from a fair response, as it can significantly alter their outputs, leading to skewed, unfair, hallucinations and/or inaccurate results [26].”
Rath, K., Cassim, F., & Haese, A. (2025) · Nordic design research conference · doi:10.21606/nordes.2025.30
Mandava, R. (2025) · International Journal of Advances in Artificial Intelligence and Machine Learning · doi:10.58723/ijaaiml.v2i2.411 · Cited by 1
Generative AI for Mental Health Education: A Comparison of ChatGPT and DeepSeek
Mu, Y., & Li, Q. (2025) · Communications in computer and information science · doi:10.1007/978-981-95-0880-8_34
Nguyen-Viet, K. (2025) · Ethics and Information Technology · doi:10.1007/s10676-025-09852-3 · Cited by 4
“Yet, AI communication tools struggle with these socio-pragmatic nuances, as their Western-centric training data prioritize syntactic accuracy based on cultural and situational cues, ensuring respectful interactions (Gallegos et al., 2024).”
InfluentialCoglianese, C., & Crum, C. R. (2025) · Oxford University Press eBooks · doi:10.1093/oxfordhb/9780198940272.013.0011 · Cited by 5
Towards Sustainable Image Synthesis: A Comprehensive Review of Text-to-Image Generation Models
Bharne, S., Sapkale, P., Sarda, E., Padiya, P., & Salunkhe, S. (2025) · International Research Journal of Multidisciplinary Technovation · doi:10.54392/irjmt2557
“(* in figure indicates the number are increasing per year).”
Augmented intelligence should be good for medicine, if medicine is to remain good for us
Idan, D., Celi, L. A., Einav, S., & Frenkel, A. (2025) · Discover Artificial Intelligence · doi:10.1007/s44163-025-00256-2
“Within the business world, research on AI bias is still in the early stages [16], and the conservative medical world tends to lag behind the rapid business world [17].”
Beyond the Algorithm: Educational Justice for Plurilingual Students in AI-Assisted Classrooms
Khari, M. (2025) · IGI Global eBooks · doi:10.4018/979-8-3373-2670-2.ch015 · Cited by 2
Assessing an Evolutionary Search Engine for Small Language Models, Prompts, and Evaluation Metrics
Lopes, C. L. D. V., & da Silva, L. M. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-15993-9_9 · Cited by 3
“…the field is observably shifting away from evaluations based on single metrics towards a more multi-dimensional approach [14] that explicitly advocates multi-metric assessment, validating models across dimensions such as accuracy, robustness, fairness, bias [11], toxicity, and efficiency.”
Evaluation of Ethical Bias in Responses Generated by Spanish LLMs
Vega-Rueda, J. C., & Rodríguez-Garavito, C. H. (2025) · Communications in computer and information science · doi:10.1007/978-3-032-08206-0_14
Knowledge prompting: How knowledge engineers use generative AI
Koutsiana, E., Walker, J., Nwachukwu, M., Zhang, B., Meroño-Peñuela, A., & Simperl, E. P. B. (2025) · Journal of Web Semantics · doi:10.1016/j.websem.2025.100873 · Cited by 13
“A great deal of recent work on LLMs for KE emphasises that research into bias in this area is of great importance [48, 130, 9, 54, 191].”
Holst, C., Wiegräbe, F., Redecker, C., & Lohweg, V. (2025) · Springer eBooks · doi:10.1007/978-3-658-48573-3_11 · Cited by 2
Standardizing Intelligence: Aligning Generative AI for Regulatory and Operational Compliance
Imperial, J. M., Jones, M. D., & Madabushi, H. T. (2025) · SuperIntelligence - Robotics - Safety & Alignment · doi:10.70777/si.v2i5.16189 · Cited by 16
Recognizing "Conformity Bias" in Large Language Models: A New Risk for Clinical UsePreprint
Nowroozzadeh, M. H., & Salari, R. (2025) · Research Square · doi:10.1101/2025.10.19.25338293
“[8] Conformity Bias, however, represents a distinct and particularly insidious form of model vulnerability.”
Hamila, S. A., Birchill, K., Harrison, J., Cao, K., Bullock, S., Hodgson, W. C., & Leech, M. T. (2025) · doi:10.2196/preprints.86208
d’Aloisio, G., Fadahunsi, T., Marco, A. D., & Sarro, F. (2025) · Information and Software Technology · doi:10.1016/j.infsof.2025.107956 · Cited by 1
Bias Assessment of Large Language Models in Gene Editing Technology through Information Extraction
陈, 梅. (2025) · Artificial Intelligence and Robotics Research · doi:10.12677/airr.2025.146131
Decoding the Mind of Large Language Models: A Quantitative Evaluation of Ideology and Biases
Hirose, M., & Uchida, M. (2025) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn64981.2025.11228711
Application of Large Language Models in Product Management: A Systematic Literature Review
Serra, V. M., Bosch, J., & Olsson, H. H. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-12089-2_20
A review on analysis for responsive intentional assessment
Bairagya, R., Das, S., Hait, S., Dutta, S., Ghosh, S., Kumari, T., Ghosh, T., Chowdhury, S., & Roy, P. (2025) · AIP conference proceedings · doi:10.1063/5.0309353
Uncovering Cultural Biases and Stereotypes in Large Language Models
Dorleon, G., & Shujaa, S. (2025) · Lecture notes in computer science · doi:10.1007/978-981-95-4861-3_5 · Cited by 2
How LLMs Handle Cultural Bias: Reactions to Asian Minority Historical Narratives
Shujaa, S., Dorleon, G., & Tang, A. (2025) · Lecture notes in computer science · doi:10.1007/978-981-95-4861-3_3 · Cited by 5
Leveraging artificial intelligence to support the application of quality improvement in healthcare
Sorlie, C., Preston, C., Aurelio, M. G., Glynn, E., Miller, L., Purani, A., Chitewe, A., McAllister, S., Moore, J., & Shah, A. (2025) · British Journal of Healthcare Management · doi:10.12968/bjhc.2025.0027
Ethics and Bias in Natural Language Processing
Talat, Z., & Blodgett, S. L. (2025) · Elsevier eBooks · doi:10.1016/b978-0-323-95504-1.01226-6 · Cited by 1
Automated Genre Analysis: Temperature and Reliability in LLMsPreprint
Grullon‐Polanco, L. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5852779
Mitigating Bias in Large Language Models Associated with Occupational Stereotypes
Chakraborty, C., & Howard, A. (2025) · IEEE International Symposium on Technology and Society (ISTAS) · doi:10.1109/istas65609.2025.11269614
“Our mitigations operate at inference time and do not debias internal representations; they can reduce stereotyping in outputs but are not substitutes for data/model-level interventions [34].”
DEFT: Data-Efficient Fine-Tuning Through Multi-Dimensional Data Selection
Dai, S., Liu, X., & Yu, Y. (2025) · IEEE Transactions on Audio Speech and Language Processing · doi:10.1109/taslpro.2025.3642562 · Cited by 1
Corpus Linguistics and Artificial Intelligence
Sardinha, T. B. (2025) · DELTA Documentação de Estudos em Lingüística Teórica e Aplicada · doi:10.1590/1678-460x202541474063
Exacerbating Differences in Polarity: Bias Adversarial Attack on Generative Large Language Models
Gu, T., Qing, G., Hao, F., & Chang, L. (2025) · IEEE Transactions on Audio Speech and Language Processing · doi:10.1109/taslpro.2025.3645995
Political Ideology Shifts in Large Language Models
Bernardelle, P., Civelli, S., Fröhling, L., Lunardi, R., Roitero, K., & Demartini, G. (2026) · Information Processing & Management · doi:10.1016/j.ipm.2026.105124 · Cited by 14
Tapia-Mandiola, S., & Araya, R. (2025) · Lecture notes in networks and systems · doi:10.1007/978-3-032-05070-0_16
Enhancing LLM Agent Effectiveness via Reflective Multi-Agent System
Mohamed, A. H., Santos, F. A., & dos Reis, J. C. (2025) · Anais do Workshop-Escola de Sistemas de Agentes, seus Ambientes e Aplicações (WESAAC) · doi:10.5753/wesaac.2025.37536 · Cited by 1
“• Subjectivity and self-assessment bias : the quality of the self-critique depends on the LLM agent’s ability to accurately assess its performed actions, which can be subjective or biased [[Gallegos et al. 2024], [Li et al. 2024a]].”
Tang, X., Ding, Y., Yang, Z., Chen, Y., Gu, Y., Yang, W., Ju, M., Cao, X., Liu, Y., & Zhang, W. (2025) · Lecture notes in computer science · doi:10.1007/978-981-95-4969-6_16 · Cited by 2
陈, 梅. (2025) · Advances in Education · doi:10.12677/ae.2025.15122284
Self-Refining Language Models to Assess Insurance Claims
Mostert, R. W., & Altahhan, A. (2025) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn64981.2025.11228917
İş Mülakatlarında Büyük Dil Modellerinin Kullanımı
Seymen, Ö. F. (2025) · Özgür Yayınları eBooks · doi:10.58830/ozgur.pub1026.c4107
Case Study: Enhancing Customer Trust Through AI-Driven Post-Purchase Management
Kumurdjieva, M., Doukovska, L., Ghouaiel, N., Dhanani, A., & Arenas, R. (2025) · International Conference on Big Data, Knowledge and Control Systems Engineering (BdKCSE) · doi:10.1109/bdkcse67969.2025.11300489
Jamil, B., & Asim, N. (2025) · Journal of Gandhara Medical and Dental Science · doi:10.37762/jgmds.13-1.835
FAIR-MASK: Mitigating Bias in Dense Embedding Retrieval Through Dimension Reduction
Burchett, T. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-12717-4_2
Mitigating Bias in Arabic NLP: A Framework for Ethical and Inclusive Data Preparation
Al-Shammari, E. T. (2025) · IEEE/ACS International Conference on Computer Systems and Applications (AICCSA) · doi:10.1109/aiccsa66935.2025.11315271
“These metrics enable precise identification of imbalances, guiding subsequent mitigation strategies [1].”
InfluentialReverse Affordances and Reflexivity of Smart Media - Reflections on Smart Media Affordances
Du, B. (2026) · Journal of Media Ethics · doi:10.1080/23736992.2025.2610996 · Cited by 1
Trained to Perceive: Investigating Gender Bias in the Outputs of Generative AI Language Models
Gupta, V. (2026) · IGI Global eBooks · doi:10.4018/979-8-3373-5017-2.ch013
Developer Prompts in Practice: An Empirical Study of Bias, Security, and Optimization
Rzig, D. E., Paul, D. J., Pister, K. S., Henkel, J. M., & Hassan, F. (2025) · Proceedings of the ... ACM-IEEE International Symposium on Empirical Software Engineering and Measurement/International Symposium on Empirical Software Engineering and Measurement · doi:10.1109/esem64174.2025.00067 · Cited by 2
“Like other machine learning systems, LLMs inherit biases from their training data (despite the best efforts of various post-training strategies) [17], [18], [19].”
Mind the Switch: Predicting the Code-Switch Boundary
Maddali, N. L., Uma, D., & Prasad, S. S. (2025) · International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP) · doi:10.1109/isai-nlp66160.2025.11320462
Ueno, T., & Inoshita, K. (2025) · IEEE Conference on Pervasive and Intelligent Computing (PICom) · doi:10.1109/picom68402.2025.00011
“In recent years, many methods have been proposed to evaluate and mitigate social biases embedded in LLMs. Gallegos et al. [10] formulated the concepts of social bias and fairness and presented three taxonomies that comprehensively classify evaluation metrics, datasets, and mitigation techniques.”
Ethical issues with large language models
Tedesco, R. (2026) · Elsevier eBooks · doi:10.1016/b978-0-443-33592-1.00003-6
Girard, W., Xu, H., & Yan, D. (2026) · Sensors · doi:10.3390/s26020676
Kyritsopoulos, C. G., Moore, A., Cram, L., Clief, J. V., & Llewellyn, C. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6108795
Ensuring Equitable Guidance: A Context-Based Approach to Bias Detection in AI Advisors
Augello, A., Sabatucci, L., Neroni, P., Casoria, L., & Caggianese, G. (2025) · IEEE International Conference on Metrology for eXtended Reality, Artificial Intelligence and Neural Engineering (MetroXRAINE) · doi:10.1109/metroxraine66377.2025.11340506
“LLMs can exhibit various forms of social bias [9], including gender bias, where certain professions or roles are disproportionately associated with specific genders, and racial bias, which can lead to skewed representations of different ethnic groups.”
InfluentialPubMed API and LLM-Driven Hybrid Retrieval System for Biomedical Question Answering
Morillo, A., Agamez, C., Puertas, E., Martínez-Santos, J. C., & Serrano, J. (2025) · IEEE Colombian Caribbean Conference (C3) · doi:10.1109/c366505.2025.11340582
The Economics of AI Overuse: Hidden Costs and Externalities of a "Free" TechnologyPreprint
Zhong, Z., & Zhong, X. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.5948794
Racial Bias Unmasked in Portuguese Sentence Embeddings: A Comparative Study
Pereira, L., Attux, R., & Gibaut, W. (2025) · International Conference on Soft Computing & Machine Intelligence (ISCMI) · doi:10.1109/iscmi67495.2025.11358508
“Parallel audits at the output level in Portuguese reveal race-conditioned discursive asymmetries in generated narratives, which corroborates our findings [1], [9]–[13].”
The digital whisperer: deciphering topics and sentiments towards ChatGPT on X (Twitter)
Zhang, T., & Lei, L. (2026) · Online Information Review · doi:10.1108/oir-05-2024-0338
Jia, K., Chen, J., Gao, H., Zhang, Y., Dong, Q., Zhang, D., Chen, H., & Qi, J. (2025) · IEEE International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom) · doi:10.1109/trustcom66490.2025.00032
“This theory provides a robust model for predicting how human stereotypes manifest, which, given that LLMs are trained on vast amounts of human-generated text [9], [14], serves as a blueprint for the models’ own latent biases.”
Leveraging LLMs for Smart Cities Qualitative Data Analysis
Covato, E., Soomro, K., Khan, Z., Bilal, M., Cormack, R., & Green, S. (2026) · Concurrency and Computation Practice and Experience · doi:10.1002/cpe.70547
How Marketers Can Utilize Generative AI in Qualitative Research
Fares, O. H. (2026) · Springer eBooks · doi:10.1007/978-3-031-75316-9_55-1
Choi, I. (2026) · Elsevier eBooks · doi:10.1016/b978-0-443-26629-4.00138-6
Cruz, E. D. L. (2026) · IGI Global eBooks · doi:10.4018/979-8-3373-3912-2.ch013
Metadata
- Authors
- Gallegos, I. O., Rossi, R. A., Barrow, J., Tanjim, M. M., Kim, S., Dernoncourt, F., Yu, T., Zhang, R., & Ahmed, N. K.
- Affiliations
- Stanford University (US)
- Adobe Research. ryrossi@adobe.com
- Pattern Data. joe.barrow@patterndataworks.com
- Adobe Research. tanjim@adobe.com
- Adobe Research. sukim@adobe.com
- Adobe Research. dernonco@adobe.com
- Adobe Research. tyu@adobe.com
- Adobe Research. ruizhang@adobe.com
- Intel (United States) (US)
- ORCID
- Venue
- Computational Linguistics
- Publisher
- MIT Press
- Published
- 2024
- Volume
- 50
- Issue
- 3
- Pages
- 1097-1179
- ISSN
- 0891-2017 (ISSN-L)
- 1530-9312
- OpenAlex
- W4399528455
- Semantic Scholar
- bcfa73aedf1b2d1ee4f168e21298a37ac55a37f7
- Cited by
- 1,245
- References
- 280
- Citations by year
- 2026: 596 · 2025: 451 · 2024: 188
- FWCI
- 132.76
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- Locations
- Computational Linguistics (publishedVersion, cc-by-nc-nd)
- arXiv (Cornell University) (submittedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Hate Speech and Cyberbullying Detection
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- large language models, social bias, fairness, bias assessment, bias mitigation, bias metrics, data preprocessing, post-processing
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-11-02, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-02 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.