Are Vision-Language Models Safe in the Wild? A Meme-Based Benchmark Study

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lee, DongGeon, Jang, Joonwon, Jeong, Jihae, Yu, Hwanjo
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911170672197632
author Lee, DongGeon
Jang, Joonwon
Jeong, Jihae
Yu, Hwanjo
author_facet Lee, DongGeon
Jang, Joonwon
Jeong, Jihae
Yu, Hwanjo
contents Rapid deployment of vision-language models (VLMs) magnifies safety risks, yet most evaluations rely on artificial images. This study asks: How safe are current VLMs when confronted with meme images that ordinary users share? To investigate this question, we introduce MemeSafetyBench, a 50,430-instance benchmark pairing real meme images with both harmful and benign instructions. Using a comprehensive safety taxonomy and LLM-based instruction generation, we assess multiple VLMs across single and multi-turn interactions. We investigate how real-world memes influence harmful outputs, the mitigating effects of conversational context, and the relationship between model scale and safety metrics. Our findings demonstrate that VLMs are more vulnerable to meme-based harmful prompts than to synthetic or typographic images. Memes significantly increase harmful responses and decrease refusals compared to text-only inputs. Though multi-turn interactions provide partial mitigation, elevated vulnerability persists. These results highlight the need for ecologically valid evaluations and stronger safety mechanisms. MemeSafetyBench is publicly available at https://github.com/oneonlee/Meme-Safety-Bench.
format Preprint
id arxiv_https___arxiv_org_abs_2505_15389
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Are Vision-Language Models Safe in the Wild? A Meme-Based Benchmark Study
Lee, DongGeon
Jang, Joonwon
Jeong, Jihae
Yu, Hwanjo
Computation and Language
Cryptography and Security
Computer Vision and Pattern Recognition
Rapid deployment of vision-language models (VLMs) magnifies safety risks, yet most evaluations rely on artificial images. This study asks: How safe are current VLMs when confronted with meme images that ordinary users share? To investigate this question, we introduce MemeSafetyBench, a 50,430-instance benchmark pairing real meme images with both harmful and benign instructions. Using a comprehensive safety taxonomy and LLM-based instruction generation, we assess multiple VLMs across single and multi-turn interactions. We investigate how real-world memes influence harmful outputs, the mitigating effects of conversational context, and the relationship between model scale and safety metrics. Our findings demonstrate that VLMs are more vulnerable to meme-based harmful prompts than to synthetic or typographic images. Memes significantly increase harmful responses and decrease refusals compared to text-only inputs. Though multi-turn interactions provide partial mitigation, elevated vulnerability persists. These results highlight the need for ecologically valid evaluations and stronger safety mechanisms. MemeSafetyBench is publicly available at https://github.com/oneonlee/Meme-Safety-Bench.
title Are Vision-Language Models Safe in the Wild? A Meme-Based Benchmark Study
topic Computation and Language
Cryptography and Security
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2505.15389