Skip to content
[ SIGNAL VERIFIED ] INDEPENDENT CYBERCRIME JOURNALISMCLASSIFIED // THE UNDERGROUND HAS A NEWSROOM / RSS
.COM
Latest News
ALL NEWS
Newsroom

Together AI: Stored Customer Prompt Samples Copied After Repository Credential Exposure

Together AI acknowledged that an unauthorized party downloaded stored prompt samples after an AI coding agent committed an access credential to a public repository.

Newsroom editorial artwork
Global Cybercrime IncidentsPublished Nov 22, 2026 at 5:30 PM ESTReviewed 2026-11-22

IPCRIME Global Cybercrime Tracker | Reviewed November 22, 2026. Together AI confirmed the theft of a stored sample of AI inference prompts, enabled by a repository credential exposed during development. This is not an allegation that OpenRouter’s own platform was breached.

Timeline and vector

Together AI says it collected authorized prompt samples for software testing, mostly from June 24 and July 4, 8 and 14. A coding agent later included the credential for the private repository in a public GitHub commit. A third party used the leaked access to download samples and tested credentials found within. Together AI published the incident notice on November 21.

Impact and controls

The company says affected prompts concerned MiniMax M3, GLM 5.2 and Kimi 2.6 traffic. It reported no reason to believe the stolen copies were further distributed and is assessing which customers require direct notification. It says it strengthened secret scanning and made zero data retention the default for new accounts.

Evidence limits

The public notice does not disclose a count of exposed customers, a complete sample inventory or independent proof that every stored prompt contained an active secret. No global incident count should be inferred from the separate OpenRouter endpoint-eligibility changes.

Source and related coverage

Together AI official November 21 notice; IPCRIME — provider privacy failures. No private prompt data accessed or republished.

Editorial note

IPCRIME distinguishes public records and documented findings from allegations, analysis and commentary. Report corrections through our contact page.