Together AI Disclosed a 2026 Prompt-Leakage Incident Affecting Three Named Models

Together AI disclosed a 2026 incident in which an unauthorized party downloaded stored prompt samples tied to three named models; its content review remains ongoing.

Reader ratingRate this article

Loading reader rating…

Privacy-conscious voting uses a signed random browser identifier and a short-lived network hash; the rating tables contain no raw IP-address column.

Reader ratings require JavaScript; the article remains fully readable without it.

What Together AI Disclosed About Its 2026 Prompt-Leakage Incident

Together AI disclosed a security incident involving a subset of prompts submitted to its serverless inference service in June and July 2026.1 The disclosure describes a past incident, not a claim that a breach is currently active.1

Incident disclosure • Reporting cutoff: November 21, 2026, 11:53 p.m. EST / November 22, 2026, 04:53 UTC1

The bottom line

An unauthorized party downloaded stored prompt samples after a private-repository credential was exposed in a public code repository.1 Together AI says its review of the affected prompt contents remains ongoing.1

Verified facts

Affected serviceThe samples came from Together AI’s serverless inference service.1
Named modelsThe disclosure names MiniMax M3, GLM-5.2, and Kimi K2.6.1
Primary request datesTogether AI says the samples primarily came from June 24, July 4, July 8, and July 14, 2026.1

What happened

Together AI says an AI coding agent committed a credential that provided access to a private Hugging Face repository.1 The company says an unauthorized party obtained that credential during the exposure window and downloaded the stored prompt samples.1 It also says the party searched the samples for credentials to other services and tested whether those credentials were valid.1

Scope and current limits

Together AI says it has not identified evidence of additional prompt leakage beyond this incident.1 It says it has no reason to believe the data was leaked further and understands that it has been deleted.1 Those are company statements, not independent confirmation.1

Customer count is not disclosed. The current official disclosure says a subset of prompts was involved but does not state how many customers were affected.1

The provider count is not stated. The official disclosure names three models; this article does not convert that list into a provider count.1

Content review remains open. Together AI says it is still determining whether sensitive personal information or other customer data was exposed.1

Changes Together AI reports

Together AI says it strengthened controls around development workflows, credentials, and customer data.1 Its listed changes include checking all pull requests for credentials and making zero data retention the default for new accounts.1

Actions for customers

  • Customers handling sensitive prompt data should enable zero data retention in their account settings, according to Together AI.1
  • Organizations with questions about the incident or its potential impact should contact security@together.ai, the address in the disclosure.1

Sources

  1. Security event: Prompt leakage for requests in June and July of 2026 — Together AI, published November 21, 2026; rechecked November 22, 2026 UTC.

Update and correction note: This article is bounded to the reporting cutoff above. Any later material change requires a newly sourced correction or successor.