Uses WildChat
JailbreakHunter: A Visual Analytics Approach for Jailbreak Prompts Discovery From Large-Scale Human-LLM Conversational Datasets
IEEE Transactions on Visualization and Computer Graphics · First arXiv submission
Provides visual analysis at the group, conversation, and individual-turn levels to find jailbreak attempts in conversation logs. Analysts combine filtering, cluster views, conversation timelines, and comparisons with previously reported prompts; a case study uses WildChat to investigate multi-turn attempts after an initial model refusal.
Data used
An earlier WildChat release containing more than 500,000 conversations, alongside LMSYS-Chat-1M.
Findings
The WildChat case study filters for multi-turn interactions and moderation scores to narrow the search to 503 conversations. Evaluation includes two case studies and interviews with nine researchers.
Section III-C: datasets. Section VI-B: WildChat case study and filters. Section VI-C: expert evaluation. Read the source.
Dates and versions
First arXiv version; the journal article appeared online in April 2025 and in the October 2025 issue. ArXiv version history.
2025 journal article; the annotations use the July 2024 preprint.
Cite this paper
Zhihua Jin, Shiyi Liu, Haotian Li, Xun Zhao, Huamin Qu. 2025. JailbreakHunter: A Visual Analytics Approach for Jailbreak Prompts Discovery From Large-Scale Human-LLM Conversational Datasets. IEEE Transactions on Visualization and Computer Graphics.
Show BibTeX
@article{jin2025jailbreakhunter,
title = {{JailbreakHunter: A Visual Analytics Approach for Jailbreak Prompts Discovery From Large-Scale Human-LLM Conversational Datasets}},
author = {Jin, Zhihua and Liu, Shiyi and Li, Haotian and Zhao, Xun and Qu, Huamin},
year = {2025},
journal = {IEEE Transactions on Visualization and Computer Graphics},
publisher = {IEEE},
volume = {31},
number = {10},
pages = {7923--7937},
doi = {10.1109/TVCG.2025.3557568},
url = {https://doi.org/10.1109/TVCG.2025.3557568}
}