Research Reveals: Public AI Agent Trajectories Leak Over 315k Secrets

matthew_d_green · x · 2026-08-11

Researchers scraped public AI agent trajectories from GitHub and Hugging Face, successfully decoding 315,320 reasoning blocks.

Although the visible text was sanitized, the underlying reasoning processes still exposed sensitive credentials and data, including API keys, passwords, access tokens, and personal emails.

Original post →

More from coding & agent

coding & agent channel →