Truffle Security Scans 7.6 PB of HuggingFace Training Data, Finds Exposed Secrets

882542F3884314B · hn · 2026-08-02

Truffle Security scanned 7.6 petabytes of public AI training data on HuggingFace, searching for leaked secrets such as API keys and passwords. The findings highlight widespread security risks in AI training datasets, urging developers to sanitize data and protect sensitive information.

Original post →

More from Safety

Safety channel →