Elamur Research⌕
☰
home / papers / 2026 / When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily Life

When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily Life

Paper · 2026 · Xinyue Lou, Xu Jinan, Jingyi Yin et al.

Summary

Multimodal LLMs are increasingly used as everyday assistants. Their answers can steer people into risky actions even when the user has no harmful intent. Most existing safety benchmarks target jailbreaks or explicit harm, and implicit-risk sets tend to be small or artificial. The authors introduce SaLAD, a benchmark of real photographs paired with ordinary-sounding questions. In these pairs, the danger only becomes clear when image and text are read together, for example a flooded street behind a request for directions. SaLAD also contains safe counterpart cases that test for oversensitivity. Instead of rewarding refusals, its scoring checks whether a response gives a specific, informative safety warning, using LLM-as-a-judge with warning-annotated references. Across 18 MLLMs, safety on risky queries is poor. Standard safety prompts and alignment datasets bring little improvement. Models largely know the relevant safety facts in text form but fail to apply them to visual scenes. This points to cross-modal reasoning as the key gap for real-world assistant safety.

Key points

  • SaLAD contains 2,013 image–text samples from real-world sources, with no AI-generated images, spread over 10 everyday categories (e.g., Food, Home, Travel, Sports, Health, Study). It balances unsafe and oversensitivity cases, and is released at github.com/xinyuelou/SaLAD.
  • Among 18 evaluated MLLMs, the best model reaches only a 57.2% safe-response rate on unsafe queries. The average on the unsafe set is 30.65%.
  • Safety prompts and alignment data (VLGuard, SPA-VL, MIS) help on SIUO but barely move SaLAD scores. Only SPA-VL gives consistent gains, around 1% for Qwen2.5-VL. VLGuard-trained models tend toward generic refusals.
  • In a text-only multiple-choice diagnosis, models exceed 80% accuracy on the underlying safety knowledge. This exposes a knowledge-versus-application gap in visual contexts.
  • GPT-4o and GPT-4o-mini judgments agree closely with human assessment. A full evaluation costs about $3 with GPT-4o and $0.2 with GPT-4o-mini.

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, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 worksSocial Sciences · 1,772 worksSafety Research · 543 works

🧭 Topics 4Research subjects and languages the work is about.Language models & pretraining · 3,320 worksEvaluation & benchmarks · 1,220 worksVision & language · 657 worksBias, fairness & safety · 517 works

Cite

Cite as Xinyue Lou, Xu Jinan, Jingyi Yin, Xiaolong Wang, Zhaolu Kang, Liaoyouwei, Yixuan Wang, Xiangyu Shi, Fengran Mo, SU Yao and Kaiyu Huang. 2026. When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily Life. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1446

@inproceedings{lou2026when,
  title = {{When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily Life}},
  author = {Xinyue Lou, Xu Jinan, Jingyi Yin, Xiaolong Wang, Zhaolu Kang, Liaoyouwei, Yixuan Wang, Xiangyu Shi, Fengran Mo, SU Yao and Kaiyu Huang},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1446},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1446}
}

References

49 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Anthropic. 2025. Claude 3.7 sonnet: Frontier reasoning made practical. Accessed on October 1, 2025. find
  2. Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, and 1 others. 2025. Qwen2. 5-vl technical report. arXiv preprint arXiv:2502.13923. — Qwen2.5-VL Technical Report
  3. Gabriela Ben Melech Stan, Estelle Aflalo, Raanan Yehezkel Rohekar, Anahita Bhiwandiwalla, Shao-Yen Tseng, Matthew Lyle Olson, Yaniv Gurwicz, Chenfei Wu, Nan Duan, and Vasudev Lal. 2024. Lvlm-intrepret: An interpretability tool for large vision-language models. In Proceedings of the IEEE/CVF Conferen… find
  4. Yoshua Bengio, Geoffrey Hinton, Andrew Yao, Dawn Song, Pieter Abbeel, Trevor Darrell, Yuval Noah Harari, Ya-Qin Zhang, Lan Xue, Shai Shalev-Shwartz, and 1 others. 2024. Managing extreme ai risks amid rapid progress. Science, 384(6698):842– 845. find
  5. Yangyi Chen, Karan Sikka, Michael Cogswell, Heng Ji, and Ajay Divakaran. 2024a. Dress: Instructing large vision-language models to align and interact with humans via natural language feedback. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14239–14250. find
  6. Zhe Chen, Weiyun Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Erfei Cui, Jinguo Zhu, Shenglong Ye, Hao Tian, Zhaoyang Liu, and 1 others. 2024b. Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling. arXiv preprint arXiv:2412.05271.
  7. Google DeepMind. 2025. Gemini 2.5 flash. Accessed on October 1, 2025. find
  8. Yi Ding, Lijun Li, Bing Cao, and Jing Shao. 2025. Rethinking bottlenecks in safety fine-tuning of vision language models. arXiv preprint arXiv:2501.18533. — [arXiv:2501.18533](https://arxiv.org/abs/2501.18533) find
  9. Xirui Li, Hengguang Zhou, Ruochen Wang, Tianyi Zhou, Minhao Cheng, and Cho-Jui Hsieh. 2024b. Mossbench: Is your multimodal language model oversensitive to safe queries? arXiv preprint arXiv:2406.17806. — [arXiv:2406.17806](https://arxiv.org/abs/2406.17806) find
  10. Yichen Gong, Delong Ran, Jinyuan Liu, Conglei Wang, Tianshuo Cong, Anyu Wang, Sisi Duan, and Xiaoyun Wang. 2025. Figstep: Jailbreaking large vision-language models via typographic visual prompts. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 23951–23959. find
  11. Yifan Li, Hangyu Guo, Kun Zhou, Wayne Xin Zhao, and Ji-Rong Wen. 2024c. Images are achilles’ heel of alignment: Exploiting visual vulnerabilities for jailbreaking multimodal large language models. In European Conference on Computer Vision, pages 174–189. Springer. find
  12. Yunhao Gou, Kai Chen, Zhili Liu, Lanqing Hong, Hang Xu, Zhenguo Li, Dit-Yan Yeung, James T Kwok, and Yu Zhang. 2024. Eyes closed, safety on: Protecting multimodal llms via image-to-text transformation. In European Conference on Computer Vision, pages 388–404. Springer. find
  13. Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. 2024a. Llava-next: Improved reasoning, ocr, and world knowledge. find
  14. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, and 1 others. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  15. Edward J Hu, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, and 1 others. 2022. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations.
  16. Xuhao Hu, Dongrui Liu, Hao Li, Xuanjing Huang, and Jing Shao. 2024. Vlsbench: Unveiling visual leakage in multimodal safety. arXiv preprint arXiv:2411.19939.
  17. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276.
  18. Hugo Laurençon, Andrés Marafioti, Victor Sanh, and Léo Tronchon. 2024. Building and better understanding vision-language models: insights and future directions. Preprint, arXiv:2408.12637. — [arXiv:2408.12637](https://arxiv.org/abs/2408.12637) find
  19. Seongyun Lee, Geewook Kim, Jiyeon Kim, Hyunji Lee, Hoyeon Chang, Sue Hyun Park, and Minjoon Seo. 2024. How does vision-language adaptation impact the safety of vision language models? arXiv preprint arXiv:2410.07571. — [arXiv:2410.07571](https://arxiv.org/abs/2410.07571) find
  20. Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Ziwei Liu, and Chunyuan Li. 2024a. Llava-onevision: Easy visual task transfer. Preprint, arXiv:2408.03326.
  21. Chengzu Li, Wenshan Wu, Huanyu Zhang, Yan Xia, Shaoguang Mao, Li Dong, Ivan Vulić, and Furu Wei. 2025. Imagine while reasoning in space: Multimodal visualization-of-thought. arXiv preprint arXiv:2501.07542. — [arXiv:2501.07542](https://arxiv.org/abs/2501.07542) find
  22. Xin Liu, Yichen Zhu, Jindong Gu, Yunshi Lan, Chao Yang, and Yu Qiao. 2024b. Mm-safetybench: A benchmark for safety evaluation of multimodal large language models. In European Conference on Computer Vision, pages 386–403. Springer. find
  23. Zheyuan Liu, Dongwhi Kim, Yixin Wan, Xiangchi Yuan, Zhaoxuan Tan, Fengran Mo, and Meng Jiang. 2026. Mtmcs-bench: Evaluating contextual safety of multimodal large language models in multi-turn dialogues. arXiv preprint arXiv:2601.06757.
  24. Xinyue Lou, You Li, Jinan Xu, Xiangyu Shi, Chi Chen, and Kaiyu Huang. 2025. Think in safety: Unveiling and mitigating safety alignment collapse in multimodal large reasoning model. Preprint, arXiv:2505.06538. find
  25. Weidi Luo, Siyuan Ma, Xiaogeng Liu, Xiaoyu Guo, and Chaowei Xiao. 2024. Jailbreakv: A benchmark for assessing the robustness of multimodal large language models against jailbreak attacks. arXiv preprint arXiv:2404.03027. find
  26. Fanqing Meng, Lingxiao Du, Zongkai Liu, Zhixiang Zhou, Quanfeng Lu, Daocheng Fu, Tiancheng Han, Botian Shi, Wenhai Wang, Junjun He, and 1 others. 2025. Mm-eureka: Exploring the frontiers of multimodal reasoning with rule-based reinforcement learning. arXiv preprint arXiv:2503.07365. find
  27. Meta. 2024. Llama 3.2: Revolutionizing edge ai and vision with open, customizable models. find
  28. Meta AI. 2024. Meta llama 3 acceptable use policy. Accessed on October 1, 2025. find
  29. OpenAI. 2024. Usage policies - openai. Accessed on October 1, 2025. find
  30. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, and 1 others. 2022. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35:27730–27744.
  31. Georgios Pantazopoulos, Amit Parekh, Malvina Nikandrou, and Alessandro Suglia. 2024. Learning to see but forgetting to follow: Visual instruction tuning makes llms more prone to jailbreak attacks. arXiv preprint arXiv:2405.04403. model in real-world conversation. arXiv preprint arXiv:2502.11903. find
  32. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084.
  33. Jihan Yang, Shusheng Yang, Anjali W Gupta, Rilyn Han, Li Fei-Fei, and Saining Xie. 2025a. Thinking in space: How multimodal large language models see, remember, and recall spaces. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 10632–10643. find
  34. Jiayin Wang, Fengran Mo, Weizhi Ma, Peijie Sun, Min Zhang, and Jian-Yun Nie. 2024a. A user-centric multi-intent benchmark for evaluating large language models. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 3588–3612. find
  35. Yi Yang, Xiaoxuan He, Hongkun Pan, Xiyan Jiang, Yan Deng, Xingtao Yang, Haoyu Lu, Dacheng Yin, Fengyun Rao, Minfeng Zhu, and 1 others. 2025b. R1-onevision: Advancing generalized multimodal reasoning through cross-modal formalization. arXiv preprint arXiv:2503.10615. find
  36. Siyin Wang, Xingsong Ye, Qinyuan Cheng, Junwen Duan, Shimin Li, Jinlan Fu, Xipeng Qiu, and Xuanjing Huang. 2024b. Safe inputs but unsafe output: Benchmarking cross-modality safety alignment of large vision-language model. arXiv preprint arXiv:2406.15279. — [arXiv:2406.15279](https://arxiv.org/abs/2406.15279) find
  37. Yuan Yao, Tianyu Yu, Ao Zhang, Chongyi Wang, Junbo Cui, Hongji Zhu, Tianchi Cai, Haoyu Li, Weilin Zhao, Zhihui He, and 1 others. 2024. Minicpm-v: A gpt-4v level mllm on your phone. arXiv preprint arXiv:2408.01800.
  38. Weiyun Wang, Zhe Chen, Wenhai Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Jinguo Zhu, Xizhou Zhu, Lewei Lu, Yu Qiao, and 1 others. 2024c. Enhancing the reasoning ability of multimodal large language models via mixed preference optimization. arXiv preprint arXiv:2411.10442.
  39. Wenxuan Wang, Xiaoyuan Liu, Kuiyi Gao, Jen-tse Huang, Youliang Yuan, Pinjia He, Shuai Wang, and Zhaopeng Tu. 2025a. Can’t see the forest for the trees: Benchmarking multimodal safety awareness for multimodal llms. arXiv preprint arXiv:2502.11184. find
  40. Youze Wang, Wenbo Hu, Yinpeng Dong, Jing Liu, Hanwang Zhang, and Richang Hong. 2025b. Align is not enough: Multimodal universal jailbreak attack against multimodal large language models. IEEE Transactions on Circuits and Systems for Video Technology. find
  41. Zhiyu Wu, Xiaokang Chen, Zizheng Pan, Xingchao Liu, Wen Liu, Damai Dai, Huazuo Gao, Yiyang Ma, Chengyue Wu, Bingxuan Wang, Zhenda Xie, Yu Wu, Kai Hu, Jiawei Wang, Yaofeng Sun, Yukun Li, Yishi Piao, Kang Guan, Aixin Liu, and 8 others. 2024. Deepseek-vl2: Mixture-of-experts vision-language models for…
  42. Shilin Xu, Yanwei Li, Rui Yang, Tao Zhang, Yueyi Sun, Wei Chow, Linfeng Li, Hang Song, Qi Xu, Yunhai Tong, and 1 others. 2025. Mixed-r1: Unified reward perspective for reasoning capability in multimodal large language models. arXiv preprint arXiv:2505.24164. find
  43. Haochen Xue, Feilong Tang, Ming Hu, Yexin Liu, Qidong Huang, Yulong Li, Chengzhi Liu, Zhongxing Xu, Chong Zhang, Chun-Mei Feng, and 1 others. 2025. Mmrc: A large-scale benchmark for understanding multimodal large language Jiabo Ye, Haiyang Xu, Haowei Liu, Anwen Hu, Ming Yan, Qi Qian, Ji Zhang, Fei H… find
  44. Long Yuan, Fengran Mo, Kaiyu Huang, Wenjie Wang, Wangyuxuan Zhai, Xiaoyu Zhu, You Li, Jinan Xu, and Jian-Yun Nie. 2025a. Omnigeo: Towards a multimodal large language models for geospatial artificial intelligence. arXiv preprint arXiv:2503.16326. — [arXiv:2503.16326](https://arxiv.org/abs/2503.16326) find
  45. Youliang Yuan, Wenxiang Jiao, Yuejin Xie, Chihao Shen, Menghan Tian, Wenxuan Wang, Jen-tse Huang, and Pinjia He. 2025b. Towards evaluating proactive risk awareness of multimodal language models. arXiv preprint arXiv:2505.17455. — [arXiv:2505.17455](https://arxiv.org/abs/2505.17455) find
  46. Yongting Zhang, Lu Chen, Guodong Zheng, Yifeng Gao, Rui Zheng, Jinlan Fu, Zhenfei Yin, Senjie Jin, Yu Qiao, Xuanjing Huang, and 1 others. 2025. Spa-vl: A comprehensive safety preference alignment dataset for vision language models. In Proceedings of the Computer Vision and Pattern Recognition Confer… find
  47. Kaiwen Zhou, Chengzhi Liu, Xuandong Zhao, Anderson Compalas, Dawn Song, and Xin Eric Wang. 2024. Multimodal situational safety. arXiv preprint arXiv:2410.06172. — [arXiv:2410.06172](https://arxiv.org/abs/2410.06172) find
  48. Yukai Zhou, Sibei Yang, and Wenjie Wang. 2025. Beyond jailbreaks: Revealing stealthier and broader llm security risks stemming from alignment failures. arXiv preprint arXiv:2506.07402. — [arXiv:2506.07402](https://arxiv.org/abs/2506.07402) find
  49. Yongshuo Zong, Ondrej Bohdal, Tingyang Yu, Yongxin Yang, and Timothy Hospedales. 2024. Safety fine-tuning at (almost) no cost: A baseline for vision large language models. arXiv preprint arXiv:2402.02207. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/lou-2026-when-helpers-become-hazards-benchmark/