Scanning 7.6 Petabytes of HuggingFace Training Data for Secrets
We scanned every public dataset on Hugging Face for live secrets: 7.6 petabytes across 187 million files. We found 221,303 working credentials, including tokens that can push code into software millions of people install.
Summary
A comprehensive scan of Hugging Face’s vast public datasets revealed over 221,000 active credentials, including sensitive tokens that could compromise widely-used software. The findings raise significant concerns about data security in machine learning environments.
Comments (9 of ) ⌛