CrackedPDFs: Paper v1 Dataset and Reproducibility Artifacts
收藏官方服务：
资源简介：
CrackedPDFs is a paired PDF security benchmark containing 29,322 PDFs derived from 4,983 base documents and organized into 9,774 benign, confounder, and injected triplets. This paper-v1 release includes the complete PDFs, metadata, labels, frozen features, frozen evaluation splits, exact baseline metrics, checksums, and reproduction instructions. It is designed to evaluate prompt-injection defenses and representation gaps in document-ingestion systems without regenerating the PDF corpus.
提供机构：
Zenodo创建时间：
2026-08-01
