Datasetpublic

Reet1207/image-prompt-injection

Image-Based Prompt Injection Dataset Synthetic dataset for prompt injection attacks on Large Vision-Language Models (LVLMs). Team Name GitHub Ritik Sinha @Ritik1207-ind Siddhant Kumar @siddhantkumar101 Udit Dadhich @UditDadhich GitHub Repository: prompt-injection-attacks-on-LVLMS Dataset Details 4,859 labeled samples 4 attack types: typographic, structural, adversarial, metadata 4 injection goals: jailbreak, exfiltration… See the full description on the dataset page: https://huggingface.co/datasets/Reet1207/image-prompt-injection.

sourceHugging Facemitupdated 1mo agoView on Hugging Face

0likes29downloads

Dataset Card

Image-Based Prompt Injection Dataset

Synthetic dataset for prompt injection attacks on Large Vision-Language Models (LVLMs).

Team

Name	GitHub
Ritik Sinha	@Ritik1207-ind
Siddhant Kumar	@siddhantkumar101
Udit Dadhich	@UditDadhich

GitHub Repository: prompt-injection-attacks-on-LVLMS

Dataset Details

—4,859 labeled samples
—4 attack types: typographic, structural, adversarial, metadata
—4 injection goals: jailbreak, exfiltration, hijacking, social_engineering
—Scored using LLaVA-1.5-7B
—Train/Val/Test splits: 3887/485/487