Measuring Queerness: Operationalization of Gender and Sexuality in AI Benchmarks

The evaluation of LLMs is said to be in crisis. Numerous studies have been addressing the threats associated with the highly formalized method of evaluating generative AI models, known as benchmarking (Blodgett et al., 2021; Campolo, 2025; Kraft et al., 2026; Raji et al., 2021; Wallach et al., 2025). There is a broad awareness of the flaws of benchmarks, and yet, they are still widely used. One key factor may be that the developments in benchmarking are not an isolated phenomenon. Instead, they can be connected to broader societal tendencies, such as the trend toward quantification and rankings, influencing a lot of fields, one of them being science and scientific practices (Espeland & Sauder, 2007; Fourcade & Healy, 2024; Krüger, 2022; Lamont, 2012; Mennicken & Salais, 2022). Based on the assumption that technology is both shaped by and shaping social practices, AI evaluation exemplifies how social norms, values, and epistemologies become embedded in technological artifacts, even when this process remains implicit. Or as Campolo (2025) puts it: “Benchmarking demarcates perhaps less what humans and models are capable of in some abstract sense and more what they actually value; benchmarks powerfully reduce these valuations into a single numerical metric on a prediction task.” (p. 2) Processes of evaluation and measurement can be particularly harmful to queer individuals, especially when their sexual or gender identities fall outside binary or heteronormative frameworks (Keyes, 2018; Magnet, 2020). While valuable contributions have critically examined queerness in benchmarks, such as those by Devinney et al. (2022) and Hobbs (2025), there remains a lack of comprehensive research critically analyzing the conceptualization and codification of queerness within NLP benchmark infrastructures. At the Queer in AI workshop, I aim to present an ongoing research project that seeks to address this gap. Our research audits benchmarks from the past decade, examining how they operationalize gender and sexuality and categorizes them based on their developers (industry labs, academic groups, and community organizations). We employ a mixed-methods framework, combining collaborative autoethnography with systematic benchmark auditing. References Blodgett, S. L., Lopez, G., Olteanu, A., Sim, R., & Wallach, H. (2021). Stereo- typing Norwegian Salmon: An Inventory of Pitfalls in Fairness Bench- mark Datasets. Proceedings of the 59th Annual Meeting of the Associ- ation for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), 1004–1015. https://doi.org/10.18653/v1/2021.acl-long.81 Campolo, A. (2025). State-of-the-Art: The Temporal Order of Benchmarking Culture. Digital Society, 4 (2), 35. https://doi.org/10.1007/s44206-025- 00190-x Devinney, H., Bj¨orklund, J., & Bj¨orklund, H. (2022). Theories of “Gender” in NLP Bias Research. 2022 ACM Conference on Fairness Accountabil- ity and Transparency, 2083–2102. Espeland, W. N., & Sauder, M. (2007). Rankings and Reactivity: How Pub- lic Measures Recreate Social Worlds. American Journal of Sociology, 113 (1), 1–40. https://doi.org/10.1086/517897 Fourcade, M., & Healy, K. (2024). The ordinal society. Harvard university press. Hobbs, J. (2025). Theories of ”Sexuality” in Natural Language Processing Bias Research. arXiv: 2506.22481 [cs.CY]. https://doi.org/10.48550/arXiv. 2506.22481 Keyes, O. (2018). The misgendering machines: Trans/hci implications of auto- matic gender recognition. Proceedings of the ACM on human-computer interaction, 2 (CSCW), 1–22. Kraft, A., Simon, J., & Schimmler, S. (2026). Social Bias in Popular Question- Answering Benchmarks. arXiv: 2505.15553 [cs.CL]. https://doi.org/ 10.48550/arXiv.2505.15553 Krüger, A. K. (2022). Soziologie des Wertens und Bewertens. transcript. Lamont, M. (2012). Toward a Comparative Sociology of Valuation and Evalua- tion. Annual Review of Sociology, 38 (1), 201–221. https://doi.org/10. 1146/annurev-soc-070308-120022 Magnet, S. A. (2020). When biometrics fail: Gender, race, and the technology of identity. Duke University Press. Mennicken, A., & Salais, R. (Eds.). (2022). The New Politics of Numbers: Utopia, Evidence and Democracy. Springer International Publishing. Raji, I. D., Bender, E. M., Paullada, A., Denton, E., & Hanna, A. (2021). AI and the Everything in the Whole Wide World Benchmark. arXiv: 2111.15366 [cs.LG]. https://doi.org/10.48550/arXiv.2111.15366 Wallach, H., Desai, M., Cooper, A. F., Wang, A., Atalla, C., Barocas, S., Blod- gett, S. L., Chouldechova, A., Corvi, E., Dow, P. A., Garcia-Gathright, J., Olteanu, A., Pangakis, N., Reed, S., Sheng, E., Vann, D., Vaughan, J. W., Vogel, M., Washington, H., & Jacobs, A. Z. (2025). Position: Evaluating Generative AI Systems Is a Social Science Measurement Challenge (2). https://doi.org/10.48550/ARXIV.2502.00561