Elamur Research⌕
☰
home / papers / 2026 / KZ-SafetyPrompts: A Kazakh Safety Evaluation Prompt Dataset for Large Language Models

KZ-SafetyPrompts: A Kazakh Safety Evaluation Prompt Dataset for Large Language Models

Summary

Tools for testing how large language models handle harmful requests rarely cover Kazakh, and model safety behaviour is known to shift between languages. To address this, the authors built KZ-SafetyPrompts, a collection of 5,717 prompts that native speakers composed directly in Kazakh (Cyrillic) and then paired with English versions for cross-lingual comparison. The prompts imitate how teenagers or children might phrase questions. They state a harmful intent without giving procedural detail, and some include Kazakhstan-specific vocabulary and situations. Each prompt has a single label from eleven categories, assigned with explicit rules for ambiguous cases. A six-stage pipeline covered planning, writing, limited LLM support followed by human rewriting, translation, labelling, and cleaning. The cleaning stage included deduplication and schema checks. The categories are mapped to established taxonomies so the data can plug into existing evaluation setups. A baseline test of GPT-4o found that it refused only a minority of prompts, and refusal rates differed widely between categories. This suggests that English-only evaluation misses safety weaknesses that appear in Kazakh.

Key points

  • Dataset contents: 5,717 natively written Kazakh prompts with English translations, stable IDs and a standardized category field. Occasional Russian mixing reflects real youth speech.
  • Eleven categories: The labels span Racist Content, Radicalization, and Regulated Goods and Illegal Activities; the youth-life topics Health, Family, and Education (Academic Pressure); and the remaining five, namely Vulgar Language, Sexual Content, Child Exploitation, Violence, and Self-Harm.
  • Taxonomy mapping: Nine categories correspond fully or partly to Llama Guard, OpenAI moderation/usage policies and HarmBench. Education and Family are youth-oriented additions with no direct counterpart in those taxonomies.
  • Baseline results: GPT-4o was tested on a stratified subset of 1,001 prompts. It refused 28.2% overall, with per-category rates ranging from 5.5% to 53.8%.
  • Role of LLMs: Models were used only for brainstorming and translation checks. Native speakers reviewed and edited every prompt, and no generated text was kept unchanged.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cited by

2 papers of our library cite it (of the 140673 whose bibliographies we read); 1 of them have a page here, 1 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/322e8tcej745

@inproceedings{zaghouani2026kz,
  title = {{KZ-SafetyPrompts: A Kazakh Safety Evaluation Prompt Dataset for Large Language Models}},
  author = {Zaghouani, Wajdi and Ibrahim, Shimaa Amer and Muratbek, Aruzhan and Zhakenov, Olzhasbek and Akhmetzhanova, Adiya},
  booktitle = {Proceedings of the SIGUL 2026 Joint Workshop with ELE, EURALI, and DCLRL: Towards Inclusivity and Equality: Language Resources and Technologies for Under-Resourced and Endangered Languages},
  year = {2026},
  doi = {10.63317/322e8tcej745},
  url = {https://doi.org/10.63317/322e8tcej745}
}

References

23 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Lora Aroyo, Alex S. Taylor, Mark Diaz, Christopher M. Homan, Alicia Parrish, Greg Serapio-Garcia, Vinodkumar Prabhakaran, and Ding Wang. 2023. Dices dataset: Diversity in conversational ai evaluation for safety. In NeurIPS 2023 Track on Datasets and Benchmarks. find
  2. Samuel Cahyawijaya, Holy Lovenia, Fajri Koto, Dea Adhista, Emmanuel Dave, Sarah Oktavianti, Salsabil Akbar, Jhonson Lee, Nuur Shadieq, Tjeng Wawan Cenggoro, Hanung Linuwih, Bryan Wilie, Galih Muridan, Genta Winata, David Moeljadi, Alham Fikri Aji, Ayu Purwarianti, and Pascale Fung. 2023. NusaWrites:… find
  3. Patrick Chao, Edoardo Debenedetti, Alexander Robey, Maksym Andriushchenko, Francesco Croce, Vikash Sehwag, Edgar Dobriban, Nicolas Flammarion, George J. Pappas, Florian Application form (Google Form) find
  4. Tramer, Hamed Hassani, and Eric Wong. 2024. Jailbreakbench: An open robustness benchmark for jailbreaking large language models. arXiv preprint arXiv:2404.01318. find our card
  5. Samuel Gehman, Suchin Gururangan, Maarten Sap, Yejin Choi, and Noah A. Smith. 2020. RealToxicityPrompts: Evaluating neural toxic degeneration in language models. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3356–3369, Online. Association for Computational Linguisti… find our card
  6. Thomas Hartvigsen, Saadia Gabriel, Hamid Palangi, Maarten Sap, Dipankar Ray, and Ece Kamar. 2022. ToxiGen: A large-scale machine-generated dataset for adversarial and implicit hate speech detection. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1:… find our card
  7. Yujia Hu, Ming Shan Hee, Preslav Nakov, and Roy Ka-Wei Lee. 2025. Toxicity red-teaming: Benchmarking LLM safety in Singapore’s low-resource languages. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 12183–12201, Suzhou, China. Association for Computat… find
  8. Hakan Inan, Kartikeya Upasani, Jianfeng Chi, Rashi Rungta, Krithika Iyer, Yuning Mao, Michael Tontchev, Qing Hu, Brian Fuller, Davide Testuggine, and Madian Khabsa. 2023. Llama guard: Llm-based input-output safeguard for human-ai conversations. arXiv preprint arXiv:2312.06674. find our card
  9. Jafar Isbarov, Arofat Akhundjanova, Mammad Hajili, Kavsar Huseynova, Dmitry Gaynullin, Anar Rzayev, Osman Tursun, Aizirek Turdubaeva, Ilshat Saetov, Rinat Kharisov, Saule Belginova, Ariana Kenbayeva, Amina Alisheva, Abdullatif Köksal, Samir Rustamov, and Duygu Ataman. 2025. TUMLU: A unified and nati… find
  10. Alyssa Lees, Vinh Q. Tran, Yi Tay, Jeffrey Sorensen, Jai Gupta, Donald Metzler, and Lucy Vasserman. 2022. A new generation of perspective api: Efficient multilingual character-level transformers. arXiv preprint arXiv:2202.11176. find
  11. Mantas Mazeika, Long Phan, Xuwang Yin, Andy Zou, Zifan Wang, Norman Mu, Elham Sakhaee, Nathaniel Li, Steven Basart, Bo Li, David Forsyth, and Dan Hendrycks. 2024. Harmbench: A standardized evaluation framework for automated red teaming and robust refusal. arXiv preprint arXiv:2402.04249. the multili… find our card
  12. OpenAI. 2025. Openai usage policies. https://openai.com/policies/ usage-policies/. Effective 2025-10-29; Accessed 2026-02-24. ↗
  13. Yuxia Wang, Haonan Li, Xudong Han, Preslav Nakov, and Timothy Baldwin. 2024b. Do-notanswer: Evaluating safeguards in LLMs. In Findings of the Association for Computational Linguistics: EACL 2024, pages 896–911, St. Julian’s, Malta. Association for Computational Linguistics. find
  14. OpenAI. 2026. Moderation guide. https://developers.openai.com/api/docs/ guides/moderation/. Accessed 2026-02-24. ↗
  15. Yicheng Xiao, Houda Bouamor, and Wajdi Zaghouani. 2024. Chinese offensive language detection: Current status and future directions. arXiv preprint arXiv:2403.18314. — [arXiv:2403.18314](https://arxiv.org/abs/2403.18314) find
  16. Ethan Perez, Saffron Huang, Francis Song, Trevor Cai, Roman Ring, John Aslanides, Amelia Glaese, Nat McAleese, and Geoffrey Irving. 2022. Red teaming language models with language models. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 3419–3448, Abu… find our card
  17. Wajdi Zaghouani, Kholoud Khalil Aldous, and Yicheng Gao. 2026. Beyond english and evasion: A human-annotated multi-domain benchmark for high-stakes llm safety evaluation in chinese. In Proceedings of the RESOURCEFUL 2026 Workshop at LREC 2026. find
  18. Prasanjit Rath, Hari Shrawgi, Parag Agrawal, and Sandipan Dandapat. 2025. LLM safety for children. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 3: Industry Track), pages 809–821, Alb… find
  19. Paul Röttger, Hannah Kirk, Bertie Vidgen, Giuseppe Attanasio, Federico Bianchi, and Dirk Hovy. 2024. XSTest: A test suite for identifying exaggerated safety behaviours in large language models. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational L… find our card
  20. Yana Veitsman and Mareike Hartmann. 2025. Recent advancements and challenges of Turkic Central Asian language processing. In Proceedings of the First Workshop on Language Models for Low-Resource Languages, pages 309–324, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics. find
  21. Wenxuan Wang, Zhaopeng Tu, Chang Chen, Youliang Yuan, Jen-tse Huang, Wenxiang Jiao, and Michael Lyu. 2024a. All languages matter: On Wajdi Zaghouani and et al. 2026. Albanianllm-safety: A safety evaluation dataset for large language models in albanian. In Proceedings of LREC 2026. find
  22. Wajdi Zaghouani, Hamdy Mubarak, and Md Rafiul Biswas. 2024. So hateful! building a multi-label hate speech annotated arabic dataset. In Proceedings of the Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). find
  23. Zhexin Zhang, Leqi Lei, Lindong Wu, Rui Sun, Yongkang Huang, Chong Long, Xiao Liu, Xuanyu Lei, Jie Tang, and Minlie Huang. 2024. SafetyBench: Evaluating the safety of large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pap… find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zaghouani-2026-kz-safetyprompts-kazakh-safety-evaluation/