Lune

EMNLP Findings2025

Self-adaptive Dataset Construction for Real-World Multimodal Safety Scenarios

Jingen Qu, Lijun Li, Bo Zhang, Yichen Yan, Jing Shao

Abstract

Multimodal large language models (MLLMs) are rapidly evolving, presenting increasingly complex safety challenges. However, current dataset construction methods, which are riskoriented, fail to cover the growing complexity of real-world multimodal safety scenarios (RMS). And due to the lack of a unified evaluation metric, their overall effectiveness remains unproven. This paper introduces a novel image-oriented self-adaptive dataset construction method for RMS, which starts with images and end constructing paired text and guidance responses. Using the image-oriented method, we automatically generate an RMS dataset comprising 35k image-text pairs with guidance responses. Additionally, we introduce a standardized safety dataset evaluation metric: finetuning a safety judge model and evaluating its capabilities on other safety datasets. Extensive experiments on various tasks demonstrate the effectiveness of the proposed image-oriented pipeline. The results confirm the scalability and effectiveness of the image-oriented approach, offering a new perspective for the construction of real-world multimodal safety datasets.

Warning: this paper contains example data that may be latently offensive or harmful.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Models cites On Evaluating Adversarial Robustness of Large Vision-Language Models
  • FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts cites On Evaluating Adversarial Robustness of Large Vision-Language Models
  • FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts cites Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Models
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on13

  1. Jailbroken: How Does LLM Safety Training Fail?Alexander Wei, Nika Haghtalab, Jacob SteinhardtNeurIPS 2023 · 2,230 citations
  2. On Evaluating Adversarial Robustness of Large Vision-Language ModelsYunqing Zhao, Tianyu Pang, Chao Du, Xiao Yang et al.NeurIPS 2023 · 404 citations
  3. FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual PromptsYichen Gong, Delong Ran, Jinyuan Liu, Conglei Wang et al.AAAI 2025 · 350 citations
  4. Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language ModelsErfan Shayegani, Yue Dong, Nael B. Abu-GhazalehICLR 2024 · 271 citations
  5. Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language ModelsYongshuo Zong, Ondrej Bohdal, Tingyang Yu, Yongxin Yang et al.ICML 2024 · 140 citations

Related papers

  1. When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily LifeXinyue Lou, Jinan Xu, Jingyi Yin, Xiaolong Wang et al.ACL Findings 2026 · 1 citation
  2. OutSafe-Bench: A Benchmark for Multimodal Offensive Content Detection in Large Language ModelsYuping Yan, Yuhan Xie, Yuanshuai Li, Yingchao Yu, Lingjuan Lyu, Yaochu JinCVPR Findings 2026 · 2 citations
  3. Multimodal Situational SafetyKaiwen Zhou, Chengzhi Liu, Xuandong Zhao, Anderson Compalas et al.ICLR 2025
  4. USB: A Comprehensive and Unified Safety Evaluation Benchmark for Multimodal Large Language ModelsBaolin Zheng, Guanlin Chen, Qingyang Teng, Hongqiong Zhong et al.ACL 2026 · 10 citations
  5. MTMCS-Bench: Evaluating Contextual Safety of Multimodal Large Language Models in Multi-Turn DialoguesZheyuan Liu, Dongwhi Kim, Yixin Wan, Xiangchi Yuan et al.ACL Findings 2026
Dusk over the sea between two cliffs drawn in fine vertical lines