Persistent Anti-Muslim Bias in Large Language Models
Open accessThe DOI matches its Crossref record; 13 of the 13 references it cites were found in a registry or an index.
Abstract
It has been observed that large-scale language models capture undesirable societal biases, e.g. relating to race and gender; yet religious bias has been relatively unexplored. We demonstrate that GPT-3, a state-of-the-art contextual language model, captures persistent Muslim-violence bias. We probe GPT-3 in various ways, including prompt completion, analogical reasoning, and story generation, to understand this anti-Muslim bias, demonstrating that it appears consistently and creatively in different uses of the model and that it is severe even compared to biases about other religious groups. For instance, Muslim is analogized to terrorist in 23% of test cases, while Jewish is mapped to its most common stereotype, money, in 5% of test cases. We quantify the positive distraction needed to overcome this bias with adversarial text prompts, and find that use of the most positive 6 adjectives reduces violent completions for Muslims from 66% to 20%, but which is still higher than for other religious groups.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
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
“Pre-trained on massive amounts of text on the Internet, LMs are known to produce linguistically fawed, factually incorrect, or even ethically problematic text at times [1,24].”
From CoAuthor: Designing a Human-AI Collaborative Writing Dataset for Exploring Language Model Capabilities (2022) · cites this paper as [1]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“LLMs and other AI tools may have a number of biases that could arise from various factors, such as unbalanced training datasets74–77.”
From On opportunities and challenges of large multimodal foundation models in education (2025) · cites this paper as [75]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“it is also a liability; all AI systems might inherit the same problematic biases of a few foundation models [Bolukbasi et al. 2016; Caliskan et al. 2017; Abid et al. 2021, inter alia]”
From On the Opportunities and Risks of Foundation Models (2021) · cites this paper as Abid et al. 2021
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Previous work has shown that language models tend to exhibit human-like biases with respect to specific protected groups such as gender, ethnicity or religious orientation3–6.”
From Generative language models exhibit social identity biases (2024) · cites this paper as [4]
In the citing paper
“If a medical LLM has been trained on data that contains misinformation or biased content, LLMs deployed for clinical decision-making are at risk of reproducing these problematic contents [120,124–126].”
From Bias in medical AI: Implications for clinical decision-making (2024) · cites this paper as [126]
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.
The self-check needs the paper’s full text.
Citations
Cites
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404 · cited 1×
“These languagemodels have evolved, from learning individual word vectors with single-layer models [8], to more complex language generation architectures such as recurrent neural networks [5] and most recently transformers [4, 6, 14].”
BackgroundEfficient Estimation of Word Representations in Vector Space
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013) · arXiv · doi:10.48550/arxiv.1301.3781 · Cited by 26,568 · cited 1×
“These languagemodels have evolved, from learning individual word vectors with single-layer models [8], to more complex language generation architectures such as recurrent neural networks [5] and most recently transformers [4, 6, 14].”
BackgroundLanguage 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 4×
“These languagemodels have evolved, from learning individual word vectors with single-layer models [8], to more complex language generation architectures such as recurrent neural networks [5] and most recently transformers [4, 6, 14].”
BackgroundMethodInfluentialSelf-Critical Sequence Training for Image Captioning
Rennie, S. J., Marcheret, E., Mroueh, Y., Ross, J., & Goel, V. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.131 · Cited by 2,112 · cited 1×
“These descriptions are generated by passing images through a pretrained image captioning model [11], and fine-tuning GPT-3 to transform the short captions into long descriptions, as shown in Fig.”
MethodReformer: The Efficient Transformer
Kitaev, N., Kaiser, Ł., & Levskaya, A. (2020) · arXiv · doi:10.48550/arxiv.2001.04451 · Cited by 1,435 · cited 1×
“These languagemodels have evolved, from learning individual word vectors with single-layer models [8], to more complex language generation architectures such as recurrent neural networks [5] and most recently transformers [4, 6, 14].”
BackgroundSemi-supervised Sequence Learning
Dai, A. M., & Le, Q. V. (2015) · arXiv · doi:10.48550/arxiv.1511.01432 · Cited by 994
StereoSet: Measuring stereotypical bias in pretrained language models
Nadeem, M., Bethke, A., & Reddy, S. (2020) · arXiv · doi:10.48550/arxiv.2004.09456 · Cited by 643 · cited 2×
“As a result, even though the various language models have different architectures, since they are trained on similar corpora of text, they often learn similar biases [9].”
BackgroundUniversal Adversarial Triggers for Attacking and Analyzing NLP
Wallace, E., Feng, S., Kandpal, N., Gardner, M., & Singh, S. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1221 · Cited by 642
Identifying and Reducing Gender Bias in Word-Level Language Models
Bordia, S., & Bowman, S. R. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Student Research Workshop · doi:10.18653/v1/n19-3002 · Cited by 219 · cited 1×
“Previous work has explored the biases that are ultimately learned by language models, with the primary focus being racial and gender stereotypes [2, 3, 7, 12].”
BackgroundThe Woman Worked as a Babysitter: On Biases in Language GenerationPreprint
Sheng, E., Chang, K.-W., Natarajan, P., & Peng, N. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.01326 · Cited by 205 · cited 2×
“In [12], the authors made similar observations about language generation models.”
Background
Research works printed but not linked: 6
- 7. T. Bolukbasi, K.-W. Chang, J. Y. Zou, V. Saligrama, A. T. Kalai, Advances in neural information processing systems 29 , 4349 (2016).
- 9. S. Bordia, S. R. Bowman, arXiv preprint arXiv:1904.03035 (2019). · arXiv
- 2. A. M. Dai, Q. V. Le, Advances in neural information processing systems 28 , 3079 (2015).
- 10. K. Lu, P. Mardziel, F. Wu, P. Amancharla, A. Datta, Logic, Language, and Security (Springer, 2020), pp. 189-202.
- 3. A. Vaswani, et al. , Advances in neural information processing systems (2017), pp. 59986008.
- 14. E. Wallace, S. Feng, N. Kandpal, M. Gardner, S. Singh, arXiv preprint arXiv:1908.07125 (2019). · arXiv
Cited by
Items 601–700 of 986
For an ethics of conversational Artificial Intelligence
Pistilli, G. (2024) · theses.fr (ABES) · doi:10.70675/621c604czc262z4d68z8f9azf5340bfca0ab · Cited by 5
González-Cantera, T., & Bonacchi, C. (2025) · International Journal of Heritage Studies · doi:10.1080/13527258.2025.2543736 · Cited by 4
“Researchers have highlighted the biased perspectives of GenAI on gender, race, class, disability, religion, and language (Abid, Farooqi, and Zou 2021).”
Context Is King: Large Language Models’ Interpretability in Divergent Knowledge Scenarios
Piñeiro-Martín, A., Santos-Criado, F.-J., García-Mateo, C., Docío-Fernández, L., & López-Pérez, M. D. C. (2025) · Applied Sciences · doi:10.3390/app15031192 · Cited by 4
Salvi, R. C., & Bosch, N. (2025) · ACM Journal on Responsible Computing · doi:10.1145/3737882 · Cited by 4
“…of interest could be sentiment or toxicity, for instance; analyzing sentiment reveals if certain groups are consistently discussed in more negative or positive tones [73], while assessing toxicity can show if some groups are more frequently associated with harmful or offensive content [1].”
Azime, I. A., Kanubala, D. D., Afonja, T., Fritz, M., Valera, I., Klakow, D., & Slusallek, P. (2025) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2025.findings-emnlp.947 · Cited by 4
Goethals, S., Rhue, L. A., & Sundararajan, A. (2025) · AI and Ethics · doi:10.1007/s43681-025-00876-5 · Cited by 5
“…information to create stereotypes [40– 42], based on the association of names with social identity, and LLMs are known to have stereotypical bias [45, 46]. agreement about the definition of “bias” [ 16], and some scholars argue that LLMs should be biased [52] to accurately reflect the language…”
Research challenges and future directions in transformer-based neural machine translation
Banerjee, A., Singh, B. K., Kumar, V., & Banik, D. (2026) · Expert Systems with Applications · doi:10.1016/j.eswa.2025.131062 · Cited by 4
Sun, W., Yin, Z., Chen, Y., Sun, W., Weng, S., & Wang, Y. (2026) · Asia Pacific Journal of Tourism Research · doi:10.1080/10941665.2026.2620378 · Cited by 4
“These biases typically reflect broader social prejudices linked to sensitive characteristics, including gender, ethnicity, age, nationality, profession, and religion (Abid et al., 2021).”
Сергеевна, З. Е., & Trapeznikov, V. P. (2026) · Полис Политические исследования · doi:10.17976/jpps/2026.01.11 · Cited by 4
Designing Social Robots with Ethical, User-Adaptive Explainability in the Era of Foundation Models
Doğan, F. I., Markelius, A., & Güneş, H. (2026) · ACM/IEEE International Conference on Human-Robot Interaction (HRI) · doi:10.1145/3776734.3794453 · Cited by 4
“Finally, persistent bias in model outputs, ranging from racialised stereotypes [3] and geopolitical distortions [53] to anti-LGBTQ+ bias [25] and gendered rationalisations [36] further reflects structural epistemic centralisation, with harmful consequences for marginalised groups [6].”
Islamophobia and anti-Muslim racism studies: review and prospect
Hopkins, P. E., & Carr, J. (2026) · Ethnic and Racial Studies · doi:10.1080/01419870.2026.2627495 · Cited by 4
Testoni, A., & Calixto, I. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-short.36 · Cited by 5
Evaluating the Effectiveness of AI Source Disclosure in Human–AI Communication
Bródy, G., Rouillard, V., Asherov, D., & Aravind, A. (2024) · doi:10.21428/e4baedd9.3f5bb369 · Cited by 4
Developing Linguistic Patterns to Mitigate Inherent Human Bias in Offensive Language Detection
Tanyel, T., Alkurdi, B., & Ayvaz, S. (2024) · TURKISH JOURNAL OF ELECTRICAL ENGINEERING & COMPUTER SCIENCES · doi:10.55730/1300-0632.4105 · Cited by 4
Guiding Reinforcement Learning Using Uncertainty-Aware Large Language Models
Shoaeinaeini, M., & Harrison, B. E. (2025) · IEEE International Conference on Trust, Privacy and Security in Intelligent Systems and Applications (TPS-ISA) · doi:10.1109/tps-isa67132.2025.00045 · Cited by 4
“Aleatoric uncertainty is related to observation errors, such as sensor noise, while Although LLMs demonstrate remarkable capabilities and rapid advancements, they frequently produce incorrect information unexpectedly, including hallucinations [42], disinformation [43], or bias [44].”
Implicit-bias-like patterns in reasoning models
Lee, M. H. J., & Lai, C. K. (2026) · Nature Machine Intelligence · doi:10.1038/s42256-026-01300-1 · Cited by 4
“We present the full set of 20 prompt variations used for data collection in Supplementary Section 2 of the Supplementary Information.”
PROBLEMS AND OPPORTUNITIES OF ARTIFICIAL INTELLIGENCE
Gürsakal, N., Çelik, S., & Batmaz, B. (2022) · Akademik Yaklaşımlar Dergisi · doi:10.54688/ayd.1104830 · Cited by 4
Kejriwal, M. (2022) · Future of business and finance · doi:10.1007/978-3-031-19039-1_3 · Cited by 4
20Q: Overlap-Free World Knowledge Benchmark for Language Models
Bruyn, M. D., Lotfi, E., Buhmann, J., & Daelemans, W. M. P. (2022) · Workshop on Generation, Evaluation and Metrics (GEM) · doi:10.18653/v1/2022.gem-1.46 · Cited by 4
“(Kejriwal et al., 2022; Kadavath et al., 2022; Lucy and Bamman, 2021; Abid et al., 2021).”
Lange, M., Koliousis, A., Fayez, F., Gogarty, E., & Twumasi, R. (2025) · PLoS ONE · doi:10.1371/journal.pone.0315768 · Cited by 4
Empirical Evaluation of Social Bias in Text Classification Systems
Djennane, L., Kardkovács, Z. T., Benatallah, B., Gaci, Y., Gaaloul, W., & Farah, Z. (2024) · International Conference on Foundation and Large Language Models (FLLM) · doi:10.1109/fllm63129.2024.10852435 · Cited by 4
“Other studies have identified biases based on race [13] and religion [14], among others.”
Ladegaard, I. (2025) · Socius Sociological Research for a Dynamic World · doi:10.1177/23780231251326981 · Cited by 4
“For instance, one LLM generated terrorism-related responses to Muslim-related prompts (Abid, Farooqi, and Zou 2021).”
Goethals, S., Sedoc, J., & Provost, F. J. (2025) · ACM Conference on User Modeling, Adaptation and Personalization (UMAP) · doi:10.1145/3708319.3733656 · Cited by 4
“For example, the characteristics of interest could include whether the outputs are biased, whether they are fabricated, whether they contain hate speech, or whether they violate policies or otherwise require content moderation actions [1, 21, 22, 30].”
EnronSR: A Benchmark for Evaluating AI-Generated Email Replies
Shay, M., Davidson, R., & Grinberg, N. (2024) · Proceedings of the International AAAI Conference on Web and Social Media · doi:10.1609/icwsm.v18i1.31448 · Cited by 5
Tambwekar, P., Dodeja, L., Vaska, N., Xu, W., & Gombolay, M. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.853 · Cited by 4
“Many sets prior work have studied approaches to identify and mitigate bias (Abid et al., 2021; Silva et al., 2021b; Guo et al., 2022; Viswanath and Zhang, 2023).”
BackgroundFisher, J. A. (2024) · Tradition Innovations in Arts Design and Media Higher Education · doi:10.9741/2996-4873.1002 · Cited by 4
TorchQL: A Programming Framework for Integrity Constraints in Machine Learning
Naik, A., Stein, A., Wu, Y., Naik, M., & Wong, E. (2024) · Proceedings of the ACM on Programming Languages · doi:10.1145/3649841 · Cited by 4
“…vehicles [Wakabayashi 2018], vision models performing worse on people with darker skin [Wilson et al. 2019], language models producing text containing offensive stereotypes [Abid et al. 2021], and medical diagnosis models degrading in performance when used in new hospitals [Zech et al. 2018].”
BackgroundDon't "research fast and break things": On the ethics of Computational Social SciencePreprint
Leslie, D. (2022) · arXiv (Cornell University) · doi:10.5281/zenodo.6635569 · Cited by 4
“2017) to structures of encoded prejudice like proxybased digital redlining (Cottom, 2016; Friedline, 2020) and the perpetuation of harmful stereotyping (Abid et al., 2021; Bommasani et al., 2021; Caliskan et al., 2017; Muñoz et al., 2021; Nadeem et al., 2020; Weidinger et al., 2021).”
BackgroundDon’t 'Research Fast and Break Things': On the Ethics of Computational Social SciencePreprint
Leslie, D. (2022) · SSRN Electronic Journal · doi:10.2139/ssrn.4134621 · Cited by 4
Proceedings of The ACM Collective Intelligence Conference
(2023) · doi:10.1145/3582269 · Cited by 4
ContraSolver: Self-Alignment of Language Models by Resolving Internal Preference ContradictionsPreprint
Zhang, X., Yin, X., & Wan, X. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2406.08842 · Cited by 4
“However, LLMs can generate undesirable outputs [Gehman et al., 2020, Carlini et al., 2021, Abid et al., 2021] and unfaithful content [Ji et al., 2023, Zhou et al., 2023], posing security risks in NLP applications.”
BackgroundWeQA: A Benchmark for Retrieval Augmented Generation in Wind Energy DomainPreprint
Meyur, R., Phan, H., Wagle, S., Strube, J., Halappanavar, M. M., Horawalavithana, S., Acharya, A., & Munikoti, S. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2408.11800 · Cited by 4
Evaluating Large Language Models with fmevalPreprint
Schwöbel, P., Franceschi, L., Zafar, M. B., Vasist, K., Malhotra, A., Shenhar, T., Tailor, P., Yilmaz, P., Diamond, M., & Donini, M. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2407.12872 · Cited by 4
“As a consequence, LLMs are at risk of regurgitating toxicity [11, 19], stereotypes [3, 32–34], and non-truthful outputs [24, 30].”
BackgroundA Thematic Framework for Human-AI Co-creative Writing: Writers' Experiences of the ProcessPreprint
Hansen, I. B., & Rafner, J. F. (2025) · OSF Preprints (OSF Preprints) · doi:10.31219/osf.io/fgqcz_v1 · Cited by 4
The Emergence of Social Science of Large Language ModelsPreprint
Jia, X., & Zhao, Z. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.24877 · Cited by 4
Hossain, K. A., Mahmud, J. S., Tuli, M. H., Mitra, A., Haque, S. M. T., & Sadeque, F. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.03943 · Cited by 4
Strimling, P., Krueger, J., & Karlsson, S. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2407.16886 · Cited by 4
“Similar to humans, these models are susceptible to various biases, including those related to race, gender, political ideology, and religious affiliation (Abid et al., 2021; Kirk et al., 2021; Rozado, 2023; Singh & Ramakrishnan, 2023).”
BackgroundBonil, G., Hashiguti, S. T., Silva, J., Gondim, J., Maia, H., Silva, N., Pedrini, H., & Avila, S. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.10304 · Cited by 4
“In summary, while this research builds on existing work on bias in LLMs — including both technical studies (Abid, Farooqi, and Zou 2021b; Lucy and Bamman 2021; Salinas, Haim, and Nyarko 2024) and socially grounded critiques (Araújo 2024a; Corazza, Silva, and Neves 2024) — it diverges by offering a…”
Analysis of Security Findings and Reduction of False Positives through Large Language Models
Wagner, J. (2024) · Qucosa (Saxon State and University Library Dresden) · Cited by 4
Reliable and Responsible Foundation Models: A Comprehensive Survey
Yang, X., Han, J., Bommasani, R., Luo, J., Qu, W., Zhou, W., Bibi, A., Wang, X., Yoon, J., Stengel-Eskin, E., et al. (2025) · Transactions on machine learning research · doi:10.48550/arxiv.2602.08145 · Cited by 5
Self-citationOverton Pluralistic Reinforcement Learning for Large Language ModelsPreprint
Fu, Y., Son, S., & Bogunovic, I. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.20759 · Cited by 4
“Since large language models are trained on data that may contain biases [Bender et al., 2021, Abid et al., 2021], training them to converge on a single answer risks privileging certain perspectives while excluding others.”
When AI Takes Sides on Questions of Faith: Persistent Asymmetries in AI-Mediated Faith GuidancePreprint
Israelsen, B., Carty, S., Coates, J., Fulda, N., Park, J., & Whiting, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.22975 · Cited by 4
Self-Debiasing Large Language Models: Zero-Shot Recognition and Reduction of Stereotypes
Gallegos, I. O., Rossi, R. A., Barrow, J., Tanjim, M. M., Yu, T., Deilamsalehy, H., Zhang, R., Kim, S., & Dernoncourt, F. (2024) · arXiv · doi:10.48550/arxiv.2402.01981 · Cited by 3
“…2020; Lu et al., 2022) and continuous prompt tuning (Fatemi et al., 2023; Yang et al., 2023) require additional fine-tuning, our work aligns more closely with techniques that prepend textual instructions or triggers to a prompt (Abid et al., 2021; Sheng et al., 2020; Narayanan Venkit et al., 2023).”
MethodYour Large Language Model is Secretly a Fairness Proponent and You Should Prompt it Like One
Li, T., Zhang, X., Du, C., Pang, T., Liu, Q., Guo, Q., Shen, C., & Liu, Y. (2024) · arXiv · doi:10.48550/arxiv.2402.12150 · Cited by 3
“…that there is a widespread presence of unfairness in LLMs. Research by Brown et al. (2020) highlights that GPT-3 exhibits social biases, and a study by Abid et al. (2021) reveals that GPT-3 demonstrates a disproportionately higher violent bias against Muslims compared to other religious groups.”
BackgroundLin, Z., Trogrlić, G., de Vreese, C. H., & Helberger, N. (2025) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3715275.3732065 · Cited by 3
“Religion-based bias often manifests as the association of Muslim with terrorism, reinforcing harmful narratives [1].”
Exploring Gender Bias in Large Language Models: An In-depth Dive into the German Language
Gnadt, K., Thulke, D., Kopeinik, S., & Schlüter, R. (2025) · Workshop on Gender Bias in Natural Language Processing (GeBNLP) · doi:10.18653/v1/2025.gebnlp-1.37 · Cited by 3
“For example, GPT-3 has been shown to (re)produce biased outputs concerning re-ligion, specifically showing anti-Muslim sentiment (Abid et al., 2021).”
Investigating Social Biases in Multimodal LLMs
Perera, M. V., Narayan, K., & Patel, V. M. (2025) · International Conference on Automatic Face and Gesture Recognition · doi:10.1109/fg61629.2025.11099161 · Cited by 3
“Additionally, Abid et al. [2] showed that GPT-3 frequently associated certain racial groups with violence.”
Generative AI in Contemporary Islamic Religious Education Discourse: Revelation Meets Algorithm
Abubakari, M. S. (2025) · IGI Global eBooks · doi:10.4018/979-8-3373-3830-9.ch007 · Cited by 3
A framework for evidence-based psychotherapy with AI (EBP-AI).
Stade, E. C., Held, P., Schwartz, H. A., Stirman, S. W., & Eichstaedt, J. C. (2026) · Journal of Psychopathology and Clinical Science · doi:10.1037/abn0001148 · Cited by 3
Developing a machine learning model to map new-build gentrification: A mixed-methods approach
Mueller, M., Quaye, I., Yi, S., Foley, J., Shah, R., Li, X., Pearsall, H., & Hoque, S. (2026) · PLoS ONE · doi:10.1371/journal.pone.0341844 · Cited by 3
Liu, Y., Lin, A., Peng, S., & Cai, Z. (2026) · Systems · doi:10.3390/systems14030286 · Cited by 3
ALIGNING AI WITH ISLAMIC VALUES: A COMPREHENSIVE SYSTEMATIC LITERATURE REVIEW ON ETHICAL HORIZONS
ABDULLAH, F., & Bashori, Y. A. (2026) · Hamdard Islamicus · doi:10.57144/hi.v49i1.1181 · Cited by 3
Characterizing User-Reported Risks across LLM Chatbots
Li, L., Ma, R., Xue, Z., & Xiong, J. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3791505 · Cited by 3
“Focuses on societal biases (race, gender) in training data and statistical disparities [1, 16, 41, 67].”
Eliza, I. J., Kamruzzaman, S., Hossen, K. M. N., Nurain, N., & Islam, A. B. M. A. A. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-29586-6_23 · Cited by 3
“On the other hand, concerns around ideological bias, cultural assumptions, and misrepresentation of religious values have also been raised [1, 30, 7].”
Kim, H., Li, X. W., Quintana, M., Biljecki, F., & Lee, S. (2026) · npj Urban Sustainability · doi:10.1038/s42949-026-00466-2 · Cited by 3
Debiasing Sentence Embedders Through Contrastive Word Pairs
Kenneweg, P., Schröder, S., Schulz, A., & Hammer, B. (2023) · International Conference on Pattern Recognition Applications and Methods · doi:10.5220/0011615300003411 · Cited by 3
“For instance, Abid, Abubakar et al. (Abid et al., 2021) showed that GPT-3 produces texts inheriting muslim-violence biases (e.g. in prompt completion).”
BackgroundLiving With Robotic Companions for Better Psychological Well-Being
Jeong, S. (2023) · XRDS Crossroads The ACM Magazine for Students · doi:10.1145/3611683 · Cited by 3
“Yet these models still suffer from hallucinations (i.e., generating false information) and biased output [12–16].”
BackgroundNegotiating becoming: a Nietzschean critique of large language models
Fischer, S. W. S., & de Boer, B. (2024) · Ethics and Information Technology · doi:10.1007/s10676-024-09783-5 · Cited by 3
“The problem is that the synthetic data generated by LLMs and other generative models are almost identical to the input data (e.g., Abid et al., 2021).”
Liu, Z., Li, G., Shi, B., Zhang, X., Ge, J., Wu, Y., & Lyu, H. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10650172 · Cited by 3
“Utilizing traditional assessment methods that involve two stages: induction and risks assessment, existing studies have demonstrated that LLMs consistently generate toxic, offensive, and biased content, even inaccurate opinions, unprofessional advice, and various other content risks [6]–[10].”
MethodEvaluating Nuanced Bias in Large Language Model Free Response Answers
Healey, J., Byrum, L., Akhtar, M. N., & Sinha, M. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-70242-6_36 · Cited by 3
GenAI and Academic Integrity: Redefining the Role of Writing Centres in a Shifting Digital Landscape
Moussu, L. (2025) · Journal of Scholarly Publishing · doi:10.3138/jsp-2024-1122 · Cited by 3
“The propagation of bias and toxicity is another pressing issue: because these systems are trained on existing data sets, they can inadvertently reproduce societal biases on race, gender, and religion, resulting in outputs that could be discriminatory (Abid et al., 2021 ; Ngo & Sakhaee, ).”
A Three-Tier LLM Framework for Forecasting Student Engagement from Qualitative Longitudinal Data
Hayat, A., Martinez, H., Khan, B., & Hasan, M. R. (2025) · Conference on Computational Natural Language Learning · doi:10.18653/v1/2025.conll-1.22 · Cited by 3
Charismatic machines: On the epistemic power of generative AI within platform convergence
Barisione, M. (2026) · New Media & Society · doi:10.1177/14614448261441417 · Cited by 3
Deutsch als Fach- und Fremdsprache: Aktuelle Tendenzen
Koch, N., & Riehl, C. M. (2024) · De Gruyter eBooks · doi:10.1515/9783110690279-010 · Cited by 3
On the Cultural Sensitivity of Large Language Models: GPT's Ability to Simulate Human Self-Concept
Luo, Y., Du, J. S., & Qiu, L. (2024) · International Conference on Behavioral, Economic, and Socio-Cultural Computing · doi:10.1109/besc64747.2024.10780488 · Cited by 3
“Our findings may address the concerns around the lack of, or biased, representation of minority human groups in AI models[42], [43], [44].”
Owls are Wise and Foxes are Unfaithful: Uncovering Animal Stereotypes in Vision Language Models
Aman, T., Nadeem, M., Sohail, S. S., Anas, M., & Cambria, E. (2025) · Lecture notes in computer science · doi:10.1007/978-981-96-8197-6_31 · Cited by 3
Von Wörtern zu Wundern: Die Technologie großer Sprachmodelle und ihre Grenzen
Simbeck, K. (2025) · Literatur - Medien - Didaktik · doi:10.57088/978-3-7329-8796-2_1 · Cited by 3
Chaffin, A., & Delaunay, J. (2023) · BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP (BlackboxNLP) · doi:10.18653/v1/2023.blackboxnlp-1.6 · Cited by 3
Chaffin, A., & Delaunay, J. (2023) · HAL · Cited by 3
Recourse for Reclamation: Chatting with Generative Language Models
Chien, J., McKee, K. R., Kay, J., & Isaac, W. S. (2024) · CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) · doi:10.1145/3613905.3650999 · Cited by 4
“For example, current models tend to produce sexist [50, 68], racist [61], casteist [49], Islamophobic [3, 68], and—more generally—otherizing content [67].”
Artificial Intelligence in Higher Education in South Africa: Some Ethical Considerations
de Villiers-Botha, T. (2024) · PhilPapers (PhilPapers Foundation) · Cited by 3
Why Algorithms Remain Unjust: Power Structures Surrounding Algorithmic ActivityPreprint
Balch, A. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2405.18461 · Cited by 4
“Algorithmic Activity is unequal in the sense that it perpetuates biases against vulnerable social groups [2, 3, 15, 38, 39, 47].”
BackgroundProbability of Differentiation Reveals Brittleness of Homogeneity Bias in GPT-4Preprint
Lee, M. H. J., & Lai, C. K. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2407.07329 · Cited by 3
“Empirical research has un-covered various biases in LLMs, ranging from negative sentiment and toxicity toward specific groups (Deshpande et al., 2023; Ousidhoum et al., 2021) to stereotypical associations (Abid et al., 2021; Nadeem et al., 2021; Lucy and Bamman, 2021).”
BackgroundLee, M. H. J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2501.02211 · Cited by 3
“We evaluate seven open-weight instruction-tuned LLMs spanning diverse architectures and parameter scales: Llama 3.1 Instruct (8B; Grattafiori et al. 2024), Qwen3 Instruct (8B; Qwen Team 2025), OLMo2 Instruct (7B; Team OLMo et al. 2024), Falcon3 Instruct (10B; Technology Innovation Institute 2024), Mistral NeMo Instruc…”
InfluentialSacred or Secular? Religious Bias in AI-Generated Financial AdvicePreprint
Khan, M. S., & Umer, H. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5193790 · Cited by 3
Focusing on Students, not Machines: Grounded Question Generation and Automated Answer GradingPreprint
Meyer, G., & Breuer, P. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.12066 · Cited by 3
DICE: A Framework for Dimensional and Contextual Evaluation of Language ModelsPreprint
Shrivastava, A., & Aoyagui, P. A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2504.10359 · Cited by 3
“While these are undeniably important to measure the adverse effects that LMs can have on society [1, 8, 44], their definitions lack granularity across different contexts.”
Augmenting Bias Detection in LLMs Using Topological Data AnalysisPreprint
Varadarajan, K., & Songdechakraiwut, T. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.07516 · Cited by 3
“Examples include the presence of gender bias in model outputs [Kirk et al., 2021, Kotek et al., 2023] and the presence of stereotypes about Muslims [Abid et al., 2021].”
A Comprehensive Study of Implicit and Explicit Biases in Large Language ModelsPreprint
Kazi, F., Young, A., Inani, Y., & Rafatirad, S. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.14153 · Cited by 3
How Implicit Bias Accumulates and Propagates in LLM Long-term MemoryPreprint
Ma, Y., Wang, L., WANG, L. Z., Yang, H., Sun, H., Xu, X., Wu, J., Chen, B., & Dong, W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.01558 · Cited by 3
“…markers (e.g., accents or religious attire) are spuriously flagged by the agent as sources of communication latency or integration risk (Gluszek & Dovidio, 2010; Abid et al., 2021), and whether non-normative family structures are penalized under the guise of stakeholder alignment (Tilcsik, 2011).”
Reward Shaping for (Inference-Time) Alignment: A Stackelberg Game PerspectivePreprint
Wang, H., Lin, T., Kong, L., Li, C., Jiang, H., & Tambe, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.02572 · Cited by 3
“The former include harmful biases – such as racial (Abid et al., 2021), gender (Kotek et al., 2023), or cultural (Li et al., 2024) bias – that motivate safety alignment.”
Large Language Models Reproduce Racial Stereotypes When Used for Text AnnotationPreprint
Törnberg, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.13891 · Cited by 3
“Prior work shows that language models encode anti-Muslim bias in text completion [1], that NLP classifiers disproportionately flag African American Vernacular English (AAVE) as toxic [5, 10, 24], and that contemporary LLMs produce racially disparate evaluations when dialect signals racial identity…”
Kilaru, S. H., Manikyala, S. T., Upadhyay, R., Ramavath, S. S. K., Nunavathu, S., & Alharthi, D. N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.12898 · Cited by 3
“The broader context is provided by foundation-model risk surveys [Bender et al., 2021] and by documentation of persistent group-targeted biases such as anti-Muslim association [Abid et al., 2021].”
Inferential Privacy Leakage in Anonymized Conversational AI LogsPreprint
Zaman, S. M. M., & Garimella, K. K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.23820 · Cited by 3
“Studies of LLM outputs have consistently shown that the models re-produce gender, religious, and racial stereotypes encoded in their training data (Sheng et al. 2019; Abid, Farooqi, and Zou 2021).”
ChatGPT in education: ethical predicaments of generative AI
Cao, Y., Fan, J., & Yang, Q. (2024) · Transactions on Social Science, Education and Humanities Research · doi:10.62051/bejkn640 · Cited by 2
“For example, generative AI has shown some religious bias, with 23% of test cases being directly compared to Muslims as terrorists [11].”
BackgroundLozoya, D. C., Liu, J., D’Alfonso, S., & Conway, M. (2024) · Workshop on Biomedical Language Processing (BioNLP) · doi:10.18653/v1/2024.bionlp-1.42 · Cited by 2
“Generative models may exhibit biases in their outputs, requiring a comprehensive examination to mitigate the inadvertent propagation of such biases (Hemmatian and Varshney, 2022; Abid et al., 2021; Cabrera Lozoya et al., 2023).”
BackgroundBakke, E. M., & Heggelund, N. W. (2025) · Fact Extraction and VERification Workshop (FEVER) · doi:10.18653/v1/2025.fever-1.12 · Cited by 2
AYA at PalmX 2025: Modeling Cultural and Islamic Knowledge in LLMs
Tajrin, J., Roy, B. B., & Alam, F. (2025) · Arabic Natural Language Processing Workshop (WANLP) · doi:10.18653/v1/2025.arabicnlp-sharedtasks.114 · Cited by 2
“Without culturally and religiously grounded priors, LLMs misinterpret idioms and taboos, amplify toxicity, and encode systematic biases, including documented anti-Muslim stereotypes (Gehman et al., 2020; Blodgett et al., 2020; Abid et al., 2021).”
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
Kieltenopetus ja tekoäly : eettisiä näkökulmia
Maijala, M., & Mäkinen, S. (2025) · Jyväskylä University Digital Archive (University of Jyväskylä) · Cited by 2
Generating the Past: How Artificial Intelligence Summaries of Historical Events Affect Knowledge
Karell, D., Shu, M., Davidson, T., & Okura, K. (2025) · Social Science Computer Review · doi:10.1177/08944393251409744 · Cited by 2
Panda, D., Anil, A., & Shukla, N. K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.06477 · Cited by 2
“Bias in NLP and Regional Bias Social biases in NLP models have primarily been studied in the context of gender (Bolukbasi et al., 2016; Sun et al., 2019), race (Dixon et al., 2018), and religion (Abid et al., 2021).”
Lamchek, J. S., & Trieu, V.-H. (2026) · Technology and Regulation · doi:10.71265/esg7cf11 · Cited by 2
Detecting doctrinal flattening in AI generated responses
Owen, J. A. (2026) · AI and Ethics · doi:10.1007/s43681-026-01051-0 · Cited by 2
“Abid, Farooqi, and Zou [11 ] documented how GPT-3 produced consistent anti-Muslim associations, while there is an array of literature that demonstrates how LLMs stereo - type religion, often flattening or misrepresenting minority traditions in persona-based generation tasks.”
When Algorithms Favor the Underrepresented – Race and Gender Biases in LLM Résumé Evaluations
Kim, S. C., Hernandez, I., & Li, J. (2026) · Journal of Personnel Psychology · doi:10.1027/1866-5888/a000388 · Cited by 2
From Expectation to Evaluation: Expectation Cues Systematically Bias LLM and Human Judgment
Sun, Y., Dillion, D., Gray, K. J., Lyu, M., Zhang, Z., & Li, F. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3790492 · Cited by 2
“From a theoretical perspective, if LLMs display expectation-aligned shifts analogous to human assimilation effects, existing psychological research may help predict when these shifts arise and how they might be mitigated [1, 13].”
The Dual Imperative in AI for OCD: Bridging Ethical Frameworks and Explainable Diagnostics
Zaboski, B. A., & Muller, G. N. (2026) · AI in Medicine · doi:10.3390/aimed1030017 · Cited by 2
“This imposes a cultural worldview rather than delivering therapeutic value [30,31].”
Li, Y., & Taneja, H. (2026) · Communication and Change · doi:10.1007/s44382-026-00030-7 · Cited by 2
“At the level of model development, pretraining on large text corpora and subsequent alignment and safety fine-tuning constrain which topics and framings are more readily produced (Abid et al., 2021; Bender et al., 2021).”
Auditing LLM Responses to Harmful Stereotypes Targeting Mental Health Groups
Dutta, A., Magu, R., Kim, S., Yoon, S., Choudhury, M. D., & KhudaBukhsh, A. R. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.2010 · Cited by 2
“…on LLM safety addresses bias, the typical focus is on identity categories such as race (Blodgett et al., 2020; Hofmann et al., 2024), religion (Abid et al., 2021; Dutta et al., 2024), gender (Wan et al., 2023), and nationality (Venkit et al., 2023), with little attention to life experiences –…”
Challenging the Myth: A Research Arc on LLMs as Human Simulacra
Münker, S., Rettinger, A., & Trilling, D. (2026) · doi:10.18653/v1/2026.bigpicture-main.4 · Cited by 2
“The finding was clear: models failed to reproduce the ideological patterns observed in human populations (Hatemi et al., 2019; Hutchinson et al., 2020; Abid et al., 2021; Liu et al., 2022).”
Metadata
- Authors
- Abid, A., Farooqi, M., & Zou, J.
- Affiliations
- Stanford University (US)
- McMaster University (CA)
- Venue
- Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society
- Publisher
- ACM
- Published
- 2021-07-21
- Pages
- 298-306
- ISSN
- 3065-8365 (ISSN-L)
- OpenAlex
- W3184144760
- Cited by
- 476
- References
- 3
- Citations by year
- 2026: 81 · 2025: 124 · 2024: 105 · 2023: 97 · 2022: 36 · 2021: 33
- FWCI
- 61.36
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- Proceedings of the AAAI/ACM Conference on AI Ethics and Society (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (cc-by)
- 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
- Islamophobia, religious bias, large language models, GPT-3, analogical reasoning, adversarial prompts, story generation, bias mitigation
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.
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.