04:00
2026-08-04
arxiv.org
large-language-models
Leak It: A Probabilistic Approach to Training-Data Extraction from Black-Box Language Models
A new study from arXiv (2608.00144v1) finds that aggregate ROC-AUC metrics hide per-document training-data leakage in black-box language models, with Pythia-6.9B reproducing exact identifiers for 16.6…