The DOI matches its Crossref record; 308 of the 308 references it cites were found in a registry or an index; 151 of the 183 sources its claims cite were checked for support.
Bias and Fairness in Large Language Models: A Survey
Open accessAbstract
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
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 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 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 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 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 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) 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)
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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
A Survey of Large Language Models
Zhao, W. X., Zhou, K., Li, J., Tang, T., Dong, Z., Hou, Y., Zhang, B., Min, Y., Zhang, J., Liu, P., et al. (2026) · Frontiers of Computer Science · doi:10.1007/s11704-026-60308-3 · Cited by 2,093
Gu, J., Jiang, X., Shi, Z., Tan, H., Zhai, X., Xu, C., Li, W., Shen, Y., Ma, S., Liu, H., et al. (2024) · arXiv · doi:10.48550/arxiv.2411.15594 · Cited by 1,781
“122 In the context of LLM-as-a-judge, it indicates that evaluators might score expressions from unfamiliar cultures poorly.”
A Survey on the Memory Mechanism of Large Language Model based Agents
Zhang, Z., Bo, X., Ma, C., Li, R., Chen, X., Dai, Q., Zhu, J., Dong, Z., & Wen, J.-R. (2024) · arXiv · doi:10.48550/arxiv.2404.13501 · Cited by 149
“Gödel Agent [206] utilizes dynamic code modification techniques to alter its own logic and behavior based on the feedback signals from the environment and stores the updated agent logic in the agent memory for recursive improvement.”
Security and Privacy Challenges of Large Language Models: A Survey
Das, B. C., Amini, M. H., & Wu, Y. (2025) · ACM Computing Surveys · doi:10.1145/3712001 · Cited by 445
A survey on LLM-based multi-agent systems: Workflow, infrastructure, and challenges
Li, X., Wang, S., Zeng, S., Wu, Y., & Yang, Y. (2024) · Vicinagearth · doi:10.1007/s44336-024-00009-2 · Cited by 404
“(3 ) Bias : Bias in LLMs manifests as the propagation and amplification of discriminatory tendencies present in training data, such as racial and gender biases, leading to unfair or harmful outputs [327–330].”
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Bias and Fairness in Large Language Models: A Survey. https://app.cochat.ai/papers/10.1162/coli_a_00524
- The paper
- Gallegos, I. O., Rossi, R. A., Barrow, J., Tanjim, M. M., Kim, S., Dernoncourt, F., Yu, T., Zhang, R., & Ahmed, N. K. (2024). Bias and Fairness in Large Language Models: A Survey. Computational Linguistics. https://doi.org/10.1162/coli_a_00524
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.