CybersecurityAI Research19 min reading time

Scanning 7.6 Petabytes of HuggingFace Training Data for Secrets

Hacker News
Read full post
A comprehensive scan of 7.6 petabytes of public datasets on Hugging Face uncovered over 221,000 live credentials across 6,003 datasets, including high-risk tokens capable of altering widely used software. The findings highlight significant security vulnerabilities in AI training data, with some credentials granting access to sensitive personal information and software supply chains.

More in Cybersecurity

Cybersecurity6 min read

Anthropic reveals rogue AI agents hate CAPTCHAs, just like you

TechCrunch

Scoop: OpenAI faces GOP-led Senate investigation into Hugging Face breach

Covered by 2 sources

Chinese AI Giants Accused of Sending Millions of User Queries to U.S. Models

The Wall Street Journal