Hacker News AIT2·71 pts
- An AI coding agent committing a private-repo credential to a public GitHub repo was the direct cause of the leak
- Exposed samples came from serverless inference requests to MiniMax M3, GLM-5.2 and Kimi K2.6, concentrated on June 24 and several July 2026 dates
- The attacker searched the downloaded samples for further credentials, risking downstream customer exposure
- Together AI advises customers handling sensitive prompts to enable zero data retention (ZDR)