The DOI matches its Crossref record; 25 of the 25 references it cites were found in a registry or an index.
UnsafeBench: Benchmarking Image Safety Classifiers on Real-World and AI-Generated Images
Abstract
With the advent of text-to-image models and concerns about their misuse, developers are increasingly relying on image safety classifiers to moderate their generated unsafe images. Yet, the performance of current image safety classifiers remains unknown for both real-world and AI-generated images. In this work, we propose UnsafeBench, a benchmarking framework that evaluates the effectiveness and robustness of image safety classifiers, with a particular focus on the impact of AI-generated images on their performance. First, we curate a large dataset of 10K real-world and AI-generated images that are annotated as safe or unsafe based on a set of 11 unsafe categories of images (sexual, violent, hateful, etc.). Then, we evaluate the effectiveness and robustness of five popular image safety classifiers, as well as three classifiers that are powered by general-purpose visual language models. Our assessment indicates that existing image safety classifiers are not comprehensive and effective enough to mitigate the multifaceted problem of unsafe images. Also, there exists a distribution shift between real-world and AI-generated images in image qualities, styles, and layouts, leading to degraded effectiveness and robustness. Motivated by these findings, we build a comprehensive image moderation tool called PerspectiveVision, which improves the effectiveness and robustness of existing classifiers, especially on AI-generated images. UnsafeBench and PerspectiveVision can aid the research community in better understanding the landscape of image safety classification in the era of generative AI.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
High-Resolution Image Synthesis with Latent Diffusion Models
Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.01042 · Cited by 20,302
Measuring nominal scale agreement among many raters.
Fleiss, J. L. (1971) · Psychological Bulletin · doi:10.1037/h0031619 · Cited by 8,733 · cited 1×
“In this study, we use the gpt-4-vision-preview checkpoint.”
DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks
Moosavi-Dezfooli, S.-M., Fawzi, A., & Frossard, P. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.282 · Cited by 5,317 · cited 1×
“With these perturbed images, we measure the MSE and RA (metrics are introduced in Section 3.4) of seven open-source classifiers.”
InfluentialUniversal Adversarial Perturbations
Moosavi-Dezfooli, S.-M., Fawzi, A., Fawzi, O., & Frossard, P. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.17 · Cited by 2,680
Liu, H., Li, C., Wu, Q., & Lee, Y. J. (2023) · Advances in Neural Information Processing Systems 36 · doi:10.52202/075280-1516 · Cited by 2,428
LAION-5B: An open large-scale dataset for training next generation image-text models
Schuhmann, C., Beaumont, R., Vencu, R., Gordon, C., Wightman, R., Cherti, M., Coombes, T., Katta, A., Mullis, C., Wortsman, M., et al. (2022) · arXiv · doi:10.48550/arxiv.2210.08402 · Cited by 1,022
InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning
Dai, W., Li, J., Li, D., Tiong, A. M. H., Zhao, J., Wang, W., Li, B., Fung, P., & Hoi, S. C. H. (2023) · neural information processing systems · doi:10.52202/075280-2142 · Cited by 422
LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs
Schuhmann, C., Vencu, R., Beaumont, R., Kaczmarczyk, R., Mullis, C., Katta, A., Coombes, T., Jitsev, J., & Komatsuzaki, A. (2021) · arXiv · doi:10.48550/arxiv.2111.02114 · Cited by 358
Fleiss’ kappa statistic without paradoxes
Falotico, R., & Quatto, P. (2015) · Quality & Quantity · doi:10.1007/s11135-014-0003-1 · Cited by 233
Schramowski, P., Tauchmann, C., & Kersting, K. (2022) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3531146.3533192 · Cited by 217 · cited 7×
“To identify a wider range of un-safe content, Chramowski et al. [63] build Q16 using the prompt learning technique on a dataset containing morally negative/positive images.”
Influential
Research works printed but not linked: 21
- Abeba Birhane, Vinay Uday Prabhu, and Emmanuel Kahembwe. 2021. Multimodal Datasets: Misogyny, Pornography, and Malignant Stereotypes. CoRR abs/2110.01963 (2021). · arXiv
- Zhi-Yi Chin, Chieh-Ming Jiang, Ching-Chun Huang, Pin-Yu Chen, andWei-Chen Chiu. 2023. Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts. CoRR abs/2309.06135 (2023). · arXiv
- Damien L. Crone, Stefan Bode, Carsten Murawski, and Simon M. Laham. 2018. The Socio-Moral Image Database (SMID): A novel stimulus set for the study of social, moral and affective processes. PLOS One (2018).
- Ning Ding, Shengding Hu, Weilin Zhao, Yulin Chen, Zhiyuan Liu, Haitao Zheng, and Maosong Sun. 2022. OpenPrompt: An Open-source Framework for Promptlearning. In Annual Meeting of the Association for Computational Linguistics (ACL). ACL, 105--113.
- Rohit Gandikota, Joanna Materzynska, Jaden Fiotto-Kaufman, and David Bau. 2023. Erasing Concepts from Diffusion Models. CoRR abs/2303.07345 (2023). · arXiv
- Ian Goodfellow, Jonathon Shlens, and Christian Szegedy. 2015. Explaining and Harnessing Adversarial Examples. In International Conference on Learning Representations (ICLR).
- Yaru Hao, Zewen Chi, Li Dong, and Furu Wei. 2022. Optimizing Prompts for Text-to-Image Generation. CoRR abs/2212.09611 (2022). · arXiv
- Lukas Helff, Felix Friedrich, Manuel Brack, Kristian Kersting, and Patrick Schramowski. 2024. LLavaGuard: VLM-based Safeguards for Vision Dataset Curation and Safety Assessment. arXiv preprint arXiv:2406.05113 (2024). · arXiv
- Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA: Low-Rank Adaptation of Large Language Models. In International Conference on Learning Representations (ICLR).
- 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. CoRR abs/2312.06674 (2023). · arXiv
- Douwe Kiela, Hamed Firooz, Aravind Mohan, Vedanuj Goswami, Amanpreet Singh, Pratik Ringshia, and Davide Testuggine. 2020. The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memes. In Annual Conference on Neural Information Processing Systems (NeurIPS). NeurIPS, 2611--2624.
- Hannah Kirk, Bertie Vidgen, Paul Röttger, Tristan Thrush, and Scott A. Hale. 2022. Hatemoji: A Test Suite and Adversarially-Generated Dataset for Benchmarking and Detecting Emoji-Based Hate. In Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT). ACL, 1352--1368.
- Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. RoBERTa: A Robustly Optimized BERT Pretraining Approach. CoRR abs/1907.11692 (2019). · arXiv
- Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. 2018. Towards Deep Learning Models Resistant to Adversarial Attacks. In International Conference on Learning Representations (ICLR).
- MaNinareh Mehrabi, Palash Goyal, Christophe Dupuy, Qian Hu, Shalini Ghosh, Richard S. Zemel, Kai-Wei Chang, Aram Galstyan, and Rahul Gupta. 2023. FLIRT: Feedback Loop In-context Red Teaming. CoRR abs/2308.04265 (2023). · arXiv
- Yiting Qu, Xinyue Shen, Xinlei He, Michael Backes, Savvas Zannettou, and Yang Zhang. 2023. Unsafe Diffusion: On the Generation of Unsafe Images and Hateful Memes From Text-To-Image Models. In ACM SIGSAC Conference on Computer and Communications Security (CCS). ACM.
- Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision. In International Conference on Machine Learning (ICML). PMLR, 8748--8763.
- Javier Rando, Daniel Paleka, David Lindner, Lennart Heim, and Florian Tramèr. 2022. Red-Teaming the Stable Diffusion Safety Filter. CoRR abs/2210.04610 (2022). · arXiv
- Naquee Rizwan, Paramananda Bhaskar, Mithun Das, Swadhin Satyaprakash Majhi, Punyajoy Saha, and Animesh Mukherjee. 2024. Zero shot VLMs for hate meme detection: Are we there yet? CoRR abs/2402.12198 (2024). · arXiv
- Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurélien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. LLaMA: Open and Efficient Foundation Language Models. CoRR abs/2302.13971 (2023). · arXiv
- Yuchen Yang, Bo Hui, Haolin Yuan, Neil Zhenqiang Gong, and Yinzhi Cao. 2023. SneakyPrompt: Evaluating Robustness of Text-to-image Generative Models' Safety Filters. CoRR abs/2305.12082 (2023). · arXiv
Cited by
VBench++: Comprehensive and Versatile Benchmark Suite for Video Generative Models
Huang, Z., Zhang, F., Xu, X., He, Y., Yu, J., Dong, Z., Ma, Q., Chanpaisit, N., Si, C., Jiang, Y., et al. (2025) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2025.3633890 · Cited by 340
Multimodal LLMs as Customized Reward Models for Text-to-Image Generation
Zhou, S., Zhang, R., Zhu, H., Kveton, B., Zhou, Y., Gu, J., Chen, J., & Chen, C. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.01826 · Cited by 14
“LLaVA-Reward can perform efficient and memory-friendly fine-tuning via Bradley-Terry ranking loss (for paired preference data) and classification loss (for unpaired affirmative data), such as Un-safeBench [43] for safety.”
InfluentialShi, S., Yin, R., Xu, N., Chen, J., Zhou, C., Du, T., Fu, Z. F., Wang, J., Wang, Z., & Ji, S. (2026) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3770855.3817756 · Cited by 1
JailbreakDiffBench: A Comprehensive Benchmark for Jailbreaking Diffusion Models
Jin, X., Weng, Z., Guo, H., Yin, C., Cheng, S., Shen, G., & Zhang, X. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.01528
“Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore. ized assessment frameworks [51].”
Metadata
- Authors
- Qu, Y., Shen, X., Wu, Y., Backes, M., Zannettou, S., & Zhang, Y.
- Affiliations
- Helmholtz Center for Information Security (DE)
- CISPA Helmholtz Center for Information Security, Saarbrücken, Germany
- Delft University of Technology (NL)
- Venue
- Proceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security
- Publisher
- ACM
- Published
- 2025-11-19
- Pages
- 3221-3235
- ISSN
- 1543-7221 (ISSN-L)
- OpenAlex
- W4416549260
- Cited by
- 6
- References
- 6
- Citations by year
- 2026: 1 · 2025: 4
- FWCI
- 4.77
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- ACM SIGSAC Conference on Computer and Communications Security (CCS) (publishedVersion)
- Funders
- European Health and Digital Executive Agency (101057917)
- BMBF (16KISAO29K)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Hate Speech and Cyberbullying Detection
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Image Retrieval and Classification Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
- Keywords
- AI-generated images, text-to-image models, benchmarking framework, distribution shift, vision-language models, generative AI misuse
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: UnsafeBench: Benchmarking Image Safety Classifiers on Real-World and AI-Generated Images. https://app.cochat.ai/papers/10.1145/3719027.3765088
- The paper
- Qu, Y., Shen, X., Wu, Y., Backes, M., Zannettou, S., & Zhang, Y. (2025). UnsafeBench: Benchmarking Image Safety Classifiers on Real-World and AI-Generated Images. Proceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security. https://doi.org/10.1145/3719027.3765088
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.