Together AI Discloses Prompt Leakage Incident Caused by Exposed Credentials

·2026.11.22 12:07

Key point

Leaked credentials allowed an unauthorized third party to download prompt samples from MiniMax M3, GLM-5.2, and Kimi K2.6 requests made in June and July 2026.

Details

  • Together AI confirmed a security incident where prompt samples were accessed by an unauthorized third party due to credentials committed to a public GitHub repository.
  • The leaked data consists of prompt samples collected for software testing purposes, specifically from requests to MiniMax M3, GLM-5.2, and Kimi K2.6 models on June 24, July 4, July 8, and July 14, 2026.
  • The third party tested five credentials found in the samples, some of which were canary credentials related to Dragos security experiments, but no evidence indicates further data exfiltration or that the data is currently active.
  • The company is cooperating with forensic investigators and law enforcement, with a review of potential personal information exposure ongoing.
  • Mitigation measures include implementing scoped or short-lived credentials, adding credential checks to all pull requests, and enabling zero data retention (ZDR) by default for new accounts.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.