Truffle Security scanned 7.6 petabytes across 187 million files on Hugging Face, uncovering 221,303 live credentials in 6,003 datasets. Among the findings, one credential exposed 393 GB of PII affecting roughly 3.7% of the global population. The scan revealed credentials with supply chain risk, including 349 GitHub personal access tokens: 223 with full repo write, 130 able to rewrite CI workflows, 112 with admin:org, and 110 that can publish packages. Additionally, 318 Docker Hub tokens capable of pushing images were identified. These credentials pose significant risk as they can alter software used by millions.

The scan also uncovered live keys unlocking real infrastructure: 8,557 GCP service-account keys across 3,811 projects, 8,594 live database logins, and 5,885 Slack and Mailgun keys. Verification was limited to metadata-only checks without accessing sensitive data. Notable exposures included 51.7 TB in non-public S3 buckets and databases exceeding 1 GB, including one MongoDB cluster at 617.7 GB.

A unique leak path involved 18 copies of an AWS key pasted into a chatbot conversation captured by LMSYS-Chat-1M. The risk is quantified at a conservative $920,000 annual exposure from stolen AI inference, based on verified keys to providers like OpenAI, Anthropic, and others. Hugging Face partnered closely with Truffle Security, and CTO Julien Chaumond contributed storage-bucket scanning support to Truffle Hog.