Skip to content
NewA calmer, faster ZeroHour

Together AI discloses prompt sample leak after exposed repository credential

Original title: Together.ai: Prompt Leakage for Requests

Hacker News · securityby Fe2O3 ()2 min read1
AI summary · x-ai/grok-4.7Summaries by Vani AI

An AI coding agent committed a Hugging Face credential; an outsider downloaded June and July 2026 prompts and tested embedded secrets, including five Dragos canaries.

Together AI notified customers that an unauthorized party downloaded prompt samples stored from its serverless inference service. Samples from MiniMax M3, GLM-5.2, and Kimi K2.6, collected for authorized testing, sat in a private Hugging Face repository. While developing the public GitHub repository xorl, an AI coding agent committed a credential; scanning caught it and the commit was reverted, but the party had already downloaded the samples and tested credentials found inside them. Together AI says it has not found further leakage; review of whether personal or other customer data was exposed is ongoing.

Unconfirmed: Whether sensitive personal information or other customer data was in the prompts is still under review.

Key facts
  • Samples mainly came from requests on June 24, July 4, July 8, and July 14, 2026.
  • Five tested secrets were canary credentials tied to a Dragos security experiment.
  • Together AI engaged an external forensic firm and law enforcement.
  • New accounts now default to zero data retention; pull requests are scanned for credentials.
  • Customers with sensitive prompts should enable zero data retention and may email security@together.ai.
Security assessment · AI, from the article

Impact: Unauthorized download of stored prompts and testing of credentials inside them

ProductAffected
Together AI serverless inference service—
What to do
  1. 1.Enable zero data retention if handling sensitive prompt data.
  2. 2.Contact security@together.ai with questions about organizational impact.
Analyze this story with Vani AIImpact, who is affected and what defenders should do, in a new chat

From the original article

2 min read at together.ai

We are notifying customers about a security incident involving a subset of prompts submitted to Together AI’s serverless inference service in June and July 2026. An unauthorized party downloaded stored prompt samples after a credential for a private repository was exposed in a public code repository. We apologize for this incident.

Hacker News · security

By Fe2O3 · together.ai

Read the full article at Hacker News · security

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.