{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,5,30]],"date-time":"2026-05-30T13:03:02Z","timestamp":1780146182842,"version":"3.54.0"},"reference-count":50,"publisher":"Association for Computing Machinery (ACM)","issue":"2","license":[{"start":{"date-parts":[[2026,5,30]],"date-time":"2026-05-30T00:00:00Z","timestamp":1780099200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/legalcode"}],"funder":[{"name":"Qatar Research Development and Innovation Council","award":["# ARG01-0525-230348"],"award-info":[{"award-number":["# ARG01-0525-230348"]}]}],"content-domain":{"domain":["dl.acm.org"],"crossmark-restriction":true},"short-container-title":["Trans. Soc. Comput."],"published-print":{"date-parts":[[2026,6,30]]},"abstract":"<jats:p>As large language models (LLMs) become deeply embedded in daily life, the urgent need for safer moderation systems that distinguish between naive and harmful requests while upholding appropriate censorship boundaries has never been greater. While existing LLMs can detect dangerous or unsafe content, they often struggle with nuanced cases such as implicit offensiveness, subtle gender and racial biases, and jailbreak prompts, due to the subjective and context-dependent nature of these issues. Furthermore, their heavy reliance on training data can reinforce societal biases, resulting in inconsistent and ethically problematic outputs. To address these challenges, we introduce GuardEval, a unified multi-perspective benchmark dataset designed for both training and evaluation, containing 106 fine-grained categories spanning human emotions, offensive and hateful language, gender and racial bias, and broader safety concerns. We also present GemmaGuard (GGuard), a Quantized Low-Rank Adaptation (QLoRA), fine-tuned version of Gemma3-12B trained on GuardEval, to assess content moderation with fine-grained labels. Our evaluation shows that GGuard achieves a macro F1 score of 0.832, substantially outperforming leading moderation models, including OpenAI Moderator (0.64) and Llama Guard (0.61). We show that multi-perspective, human-centered safety benchmarks are critical for mitigating inconsistent moderation decisions. GuardEval and GGuard together demonstrate that diverse, representative data materially improve safety, and adversarial robustness on complex, borderline cases.<\/jats:p>","DOI":"10.1145\/3815159","type":"journal-article","created":{"date-parts":[[2026,5,6]],"date-time":"2026-05-06T11:03:56Z","timestamp":1778065436000},"page":"1-32","update-policy":"https:\/\/doi.org\/10.1145\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["A Multi-Perspective Benchmark Dataset and Moderation Model for LLM Safety Evaluation with Adversarial Robustness Analysis"],"prefix":"10.1145","volume":"9","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-1865-1800","authenticated-orcid":false,"given":"Naseem","family":"Machlovi","sequence":"first","affiliation":[{"name":"Computer and Information Science, Fordham University - Rose Hill Campus","place":["Bronx, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-1021-5078","authenticated-orcid":false,"given":"Maryam","family":"Saleki","sequence":"additional","affiliation":[{"name":"Computer and Information Science, Fordham University - Rose Hill Campus","place":["Bronx, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6540-3415","authenticated-orcid":false,"given":"Ruhul","family":"Amin","sequence":"additional","affiliation":[{"name":"Computer and Information Science, Fordham University","place":["New York, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9701-5505","authenticated-orcid":false,"given":"Mohamed","family":"Rahouti","sequence":"additional","affiliation":[{"name":"Computer and Information Science, Fordham University","place":["New York, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6738-2352","authenticated-orcid":false,"given":"Shawqi","family":"Al-Maliki","sequence":"additional","affiliation":[{"name":"Information and Computing Technology, Hamad Bin Khalifa University","place":["Doha, Qatar"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9466-2475","authenticated-orcid":false,"given":"Junaid","family":"Qadir","sequence":"additional","affiliation":[{"name":"Computer Science and Engineering, Qatar University","place":["Doha, Qatar"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3261-7588","authenticated-orcid":false,"given":"Mohamed","family":"Abdallah","sequence":"additional","affiliation":[{"name":"College of Science and Engineering, Hamad Bin Khalifa University","place":["Doha, Qatar"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-0903-1204","authenticated-orcid":false,"given":"Ala","family":"Al-Fuqaha","sequence":"additional","affiliation":[{"name":"College of Science and Engineering, Hamad Bin Khalifa University","place":["Doha, Qatar"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"320","published-online":{"date-parts":[[2026,5,30]]},"reference":[{"key":"e_1_3_4_2_2","unstructured":"Marah Abdin Jyoti Aneja Harkirat Behl S\u00e9bastien Bubeck Ronen Eldan Suriya Gunasekar Michael Harrison Russell J. Hewett Mojan Javaheripi Piero Kauffmann et\u00a0al. 2024. Phi-4 technical report. (2024)."},{"key":"e_1_3_4_3_2","doi-asserted-by":"crossref","unstructured":"NIST AI. 2023. Artificial intelligence risk management framework (AI RMF 1.0). (2023) 100\u20131. https:\/\/nvlpubs.nist.gov\/nistpubs\/ai\/nist.ai","DOI":"10.6028\/NIST.AI.100-1.jpn"},{"key":"e_1_3_4_4_2","volume-title":"Proceedings of the International AAAI Conference on Web and Social Media","author":"Saha H. Tharad et\u00a0al. B. Mathew, P.","year":"2021","unstructured":"H. Tharad et\u00a0al. B. Mathew, P. Saha. 2021. Threat, abuse, and hate detection on social media: A dynamic thresholding approach. In Proceedings of the International AAAI Conference on Web and Social Media."},{"key":"e_1_3_4_5_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2021.emnlp-main.397"},{"key":"e_1_3_4_6_2","unstructured":"Rishi Bommasani Drew A. Hudson Ehsan Adeli Russ Altman Simran Arora Sydney von Arx and Michael S.2022. On the Opportunities and Risks of Foundation Models. arXiv:2108.07258. Retrieved from https:\/\/arxiv.org\/abs\/2108.07258"},{"key":"e_1_3_4_7_2","unstructured":"Faeze Brahman Sachin Kumar Vidhisha Balachandran Pradeep Dasigi Valentina Pyatkin Abhilasha Ravichander Sarah Wiegreffe Nouha Dziri Khyathi Chandu Jack Hessel et\u00a0al. 2024. The Art of Saying No: Contextual Noncompliance in Language Models. arXiv:2407.12043. Retrieved from https:\/\/arxiv.org\/abs\/2407.12043"},{"key":"e_1_3_4_8_2","doi-asserted-by":"crossref","unstructured":"Amit Das Mostafa Rahgouy Dongji Feng Zheng Zhang Tathagata Bhattacharya Nilanjana Raychawdhary Fatemeh Jamshidi Vinija Jain Aman Chadha Mary Sandage et\u00a0al. 2024. OffensiveLang: A Community Based Implicit Offensive Language Dataset.","DOI":"10.1109\/ACCESS.2024.3513220"},{"key":"e_1_3_4_9_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/N19-1423"},{"key":"e_1_3_4_10_2","volume-title":"Findings of EMNLP","author":"Camacho-Collados L. Espinosa-Anke F. Barbieri, J.","year":"2020","unstructured":"L. Espinosa-Anke F. Barbieri, J. Camacho-Collados and L. Neves. 2020. TweetEval: Unified benchmark and comparative evaluation for tweet classification. In Findings of EMNLP."},{"key":"e_1_3_4_11_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2020.emnlp-main.48"},{"key":"e_1_3_4_12_2","doi-asserted-by":"crossref","unstructured":"Samuel Gehman Suchin Gururangan Maarten Sap Yejin Choi and Noah A. Smith. 2020. RealToxicityPrompts: Evaluating neural toxic degeneration in language models.","DOI":"10.18653\/v1\/2020.findings-emnlp.301"},{"key":"e_1_3_4_13_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2025.naacl-long.306"},{"key":"e_1_3_4_14_2","doi-asserted-by":"publisher","DOI":"10.5204\/lthj.v1i0.1386"},{"key":"e_1_3_4_15_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.emnlp-main.85"},{"key":"e_1_3_4_16_2","unstructured":"Seungju Han Kavel Rao Allyson Ettinger Liwei Jiang Bill Yuchen Lin Nathan Lambert Yejin Choi and Nouha Dziri. 2024. WildGuard: Open One-Stop Moderation Tools for Safety Risks Jailbreaks and Refusals of LLMs. arXiv:2406.18495. Retrieved from https:\/\/arxiv.org\/abs\/2406.18495"},{"key":"e_1_3_4_17_2","article-title":"MedSafetyBench: Evaluating and improving the medical safety of large language models","author":"Han Tessa","year":"2024","unstructured":"Tessa Han, Aounon Kumar, Chirag Agarwal, and Himabindu Lakkaraju. 2024. MedSafetyBench: Evaluating and improving the medical safety of large language models. NeurIPS (2024).","journal-title":"NeurIPS"},{"key":"e_1_3_4_18_2","volume-title":"Proceedings of the International Conference on Learning Representations","author":"Hendrycks Dan","year":"2021","unstructured":"Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021. Measuring massive multitask language understanding. In Proceedings of the International Conference on Learning Representations."},{"key":"e_1_3_4_19_2","unstructured":"Devansh Jain Priyanshu Kumar Samuel Gehman Xuhui Zhou Thomas Hartvigsen and Maarten Sap. 2024. PolygloToxicityPrompts: Multilingual evaluation of neural toxic degeneration in large language models. arXiv:2405.09373. Retrieved from https:\/\/arxiv.org\/abs\/2405.09373"},{"key":"e_1_3_4_20_2","article-title":"Beavertails: Towards improved safety alignment of LLM via a human-preference dataset","author":"Ji Jiaming","year":"2023","unstructured":"Jiaming Ji, Mickel Liu, Josef Dai, Xuehai Pan, Chi Zhang, Ce Bian, Boyuan Chen, Ruiyang Sun, Yizhou Wang, and Yaodong Yang. 2023. Beavertails: Towards improved safety alignment of LLM via a human-preference dataset. Advances in Neural Information Processing Systems (2023).","journal-title":"Advances in Neural Information Processing Systems"},{"key":"e_1_3_4_21_2","unstructured":"Liwei Jiang Jena D. Hwang Chandra Bhagavatula Ronan Le Bras Jenny Liang Jesse Dodge Keisuke Sakaguchi Maxwell Forbes Jon Borchardt Saadia Gabriel et\u00a0al. 2021. Can machines learn morality? the delphi experiment. (2021)."},{"key":"e_1_3_4_22_2","unstructured":"Jigsaw and Google. 2018. Jigsaw Toxic Comments Dataset."},{"key":"e_1_3_4_23_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2022.emnlp-main.267"},{"key":"e_1_3_4_24_2","unstructured":"Hannah Rose Kirk Alexander Whitefield Paul R\u00f6ttger Andrew Bean Katerina Margatina Juan Ciro Rafael Mosquera Max Bartolo Adina Williams He He et\u00a0al. 2024. The PRISM Alignment Project: What Participatory Representative and Individualised Human Feedback Reveals About the Subjective and Multicultural Alignment of Large Language Models. arXiv:2404.16019. Retrieved from https:\/\/arxiv.org\/abs\/2404.16019"},{"key":"e_1_3_4_25_2","unstructured":"Xuechen Li Tianyi Zhang Yann Dubois Rohan Taori Ishaan Gulrajani Carlos Guestrin Percy Liang and Tatsunori B. Hashimoto. 2023. Alpacaeval: An Automatic Evaluator of Instruction-following Models."},{"key":"e_1_3_4_26_2","unstructured":"Percy Liang Rishi Bommasani Tony Lee Dimitris Tsipras Dilara Soylu Michihiro Yasunaga and Yian Zhang. 2023. Holistic Evaluation of Language Models. arXiv:2211.09110. Retrieved from https:\/\/arxiv.org\/abs\/2211.09110"},{"key":"e_1_3_4_27_2","doi-asserted-by":"publisher","DOI":"10.1093\/oso\/9780190905033.001.0001"},{"key":"e_1_3_4_28_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/D19-1387"},{"key":"e_1_3_4_29_2","unstructured":"AI @ Meta Llama Team. 2024. The Llama 3 Herd of Models. arXiv:2407.21783. Retrieved from https:\/\/arxiv.org\/abs\/2407.21783"},{"key":"e_1_3_4_30_2","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-032-13184-3_24"},{"key":"e_1_3_4_31_2","doi-asserted-by":"crossref","unstructured":"J. Maynez S. Narayan B. Bohnet and R. McDonald. 2020. On faithfulness and factuality in abstractive summarization. (2020).","DOI":"10.18653\/v1\/2020.acl-main.173"},{"key":"e_1_3_4_32_2","unstructured":"Meta AI. 2024. Llama-Guard 3 8B: Safety Classifier Model. Retrieved from https:\/\/huggingface.co\/meta-llama\/Llama-Guard-3-8B. Accessed: 2025-07-28."},{"key":"e_1_3_4_33_2","first-page":"928","volume-title":"Proceedings of the International Conference on Complex Networks and Their Applications","author":"Mozafari Marzieh","year":"2019","unstructured":"Marzieh Mozafari, Reza Farahbakhsh, and Noel Crespi. 2019. A BERT-based transfer learning approach for hate speech detection in online social media. In Proceedings of the International Conference on Complex Networks and Their Applications. Springer, 928\u2013940."},{"key":"e_1_3_4_34_2","unstructured":"OpenAI. 2025. OpenAI Moderation API. Retrieved from https:\/\/platform.openai.com\/docs\/guides\/moderationAccessed: 2025-01-08."},{"key":"e_1_3_4_35_2","doi-asserted-by":"publisher","DOI":"10.1609\/icwsm.v18i1.31445"},{"key":"e_1_3_4_36_2","unstructured":"Yiting Qu Xinyue Shen Yixin Wu Michael Backes Savvas Zannettou and Yang Zhang. 2024. UnsafeBench: Benchmarking Image Safety Classifiers on Real-World and AI-Generated Images. arXiv:2405.03486. Retrieved from https:\/\/arxiv.org\/abs\/2405.03486"},{"issue":"8","key":"e_1_3_4_37_2","first-page":"9","article-title":"Language models are unsupervised multitask learners","volume":"1","author":"Radford Alec","year":"2019","unstructured":"Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et\u00a0al. 2019. Language models are unsupervised multitask learners. OpenAI Blog 1, 8 (2019), 9.","journal-title":"OpenAI Blog"},{"key":"e_1_3_4_38_2","doi-asserted-by":"crossref","unstructured":"Sara Rosenthal Pepa Atanasova Georgi Karadzhov Marcos Zampieri and Preslav Nakov. 2021. SOLID: A Large-Scale Semi-Supervised Dataset for Offensive Language Identification. arXiv:2004.14454. Retrieved from https:\/\/arxiv.org\/abs\/2004.14454","DOI":"10.18653\/v1\/2021.findings-acl.80"},{"key":"e_1_3_4_39_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.naacl-long.301"},{"key":"e_1_3_4_40_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2021.acl-long.4"},{"key":"e_1_3_4_41_2","doi-asserted-by":"crossref","unstructured":"Paul R\u00f6ttger Fabio Pernisi Bertie Vidgen and Dirk Hovy. 2025. SafetyPrompts: A Systematic Review of Open Datasets for Evaluating and Improving Large Language Model Safety. arXiv:2404.05399. Retrieved from https:\/\/arxiv.org\/abs\/2404.05399","DOI":"10.1609\/aaai.v39i26.34975"},{"key":"e_1_3_4_42_2","doi-asserted-by":"publisher","DOI":"10.1093\/oso\/9780190905033.003.0017"},{"key":"e_1_3_4_43_2","doi-asserted-by":"publisher","DOI":"10.1145\/3600211.3604673"},{"key":"e_1_3_4_44_2","volume-title":"Proceedings of the ACM SIGSAC Conference on Computer and Communications Security (CCS)","author":"Shen Xinyue","year":"2024","unstructured":"Xinyue Shen, Zeyuan Chen, Michael Backes, Yun Shen, and Yang Zhang. 2024. \u201cDo Anything Now\u201d: Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Models. In Proceedings of the ACM SIGSAC Conference on Computer and Communications Security (CCS). ACM."},{"key":"e_1_3_4_45_2","doi-asserted-by":"crossref","unstructured":"Emily Sheng Kai-Wei Chang Premkumar Natarajan and Nanyun Peng. 2019. The woman worked as a babysitter: On biases in language generation. (2019).","DOI":"10.18653\/v1\/D19-1339"},{"key":"e_1_3_4_46_2","article-title":"Beyond the imitation game: Quantifying and extrapolating the capabilities of language models","author":"Srivastava Aarohi","year":"2023","unstructured":"Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao, Abu Awal Md Shoeb, Abubakar Abid, Adam Fisch, Adam R. Brown, Adam Santoro, and Aditya Gupta. 2023. Beyond the imitation game: Quantifying and extrapolating the capabilities of language models. Transactions on Machine Learning Research (2023).","journal-title":"Transactions on Machine Learning Research"},{"key":"e_1_3_4_47_2","unstructured":"Laura Weidinger John Mellor Maribeth Rauh Conor Griffin Jonathan Uesato Po-Sen Huang Myra Cheng Mia Glaese Borja Balle Atoosa Kasirzadeh et\u00a0al. 2021. Ethical and social risks of harm from language models. arXiv:2112.04359. Retrieved from https:\/\/arxiv.org\/abs\/2112.04359"},{"key":"e_1_3_4_48_2","unstructured":"Meng Ye Karan Sikka Katherine Atwell Sabit Hassan Ajay Divakaran and Malihe Alikhani. 2023. Multilingual Content Moderation: A Case Study on Reddit. arXiv:2302.09618. Retrieved from https:\/\/arxiv.org\/abs\/2302.09618"},{"key":"e_1_3_4_49_2","doi-asserted-by":"crossref","unstructured":"Marcos Zampieri Shervin Malmasi Preslav Nakov Sara Rosenthal Noura Farra and Ritesh Kumar. 2019. Predicting the type and target of offensive posts in social media. (2019).","DOI":"10.18653\/v1\/N19-1144"},{"key":"e_1_3_4_50_2","unstructured":"Wenjun Zeng Yuchi Liu Ryan Mullins Ludovic Peran Joe Fernandez Hamza Harkous Karthik Narasimhan Drew Proud Piyush Kumar Bhaktipriya Radharapu et\u00a0al. 2024. ShieldGemma: Generative AI Content Moderation Based on Gemma. arXiv:2407.21772. Retrieved from https:\/\/arxiv.org\/abs\/2407.21772"},{"key":"e_1_3_4_51_2","series-title":"NIPS \u201923","volume-title":"Proceedings of the 37th International Conference on Neural Information Processing Systems","author":"Zheng Lianmin","year":"2023","unstructured":"Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric P. Xing, et\u00a0al. 2023. Judging LLM-as-a-judge with MT-bench and Chatbot arena. In Proceedings of the 37th International Conference on Neural Information Processing Systems (New Orleans, LA, USA) (NIPS \u201923). Curran Associates Inc., Red Hook, NY, USA, Article 2020, 29 pages."}],"container-title":["ACM Transactions on Social Computing"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dl.acm.org\/doi\/pdf\/10.1145\/3815159","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,5,30]],"date-time":"2026-05-30T12:57:46Z","timestamp":1780145866000},"score":1,"resource":{"primary":{"URL":"https:\/\/dl.acm.org\/doi\/10.1145\/3815159"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,5,30]]},"references-count":50,"journal-issue":{"issue":"2","published-print":{"date-parts":[[2026,6,30]]}},"alternative-id":["10.1145\/3815159"],"URL":"https:\/\/doi.org\/10.1145\/3815159","relation":{},"ISSN":["2469-7818","2469-7826"],"issn-type":[{"value":"2469-7818","type":"print"},{"value":"2469-7826","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,5,30]]},"assertion":[{"value":"2025-09-09","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-04-21","order":2,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-05-30","order":3,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}]}}