Skip to content
VECTOR WIREAI INTELLIGENCE
UTC
Home

Together AI coding agent leaked credential, exposing customer prompts to unauthorized download

AI coding agent committed a Hugging Face credential while developing Together AI's public xorl repository.

Together AI disclosed on November 21, 2026 that an unauthorized party downloaded stored samples from a subset of customer prompts submitted to its serverless inference service1. The breach originated when an AI coding agent committed a credential granting access to a private Hugging Face Inc. repository while Together AI was developing its public xorl GitHub repository.

Together AI's automated scanning detected the exposure, and the company reverted the commit. Reverting the commit, however, did not stop the credential from being used during the exposure window. During that window, an unauthorized party obtained the credential and used it to download the stored prompt samples.

What the intruder accessed

The stored samples involved requests to the MiniMax M3, GLM-5.2, and Kimi K2.6 models, primarily on June 24, 2026, July 4, 2026, July 8, 2026, and July 14, 2026. Customers had authorized the prompts for routine software testing, and Together AI had stored the samples in a private Hugging Face repository.

After the download, the intruder searched the leaked prompts for credentials to other services and tested whether they were valid. Five of the tested credentials were canary credentials associated with a Dragos security experiment. The disclosure followed a Dragos study into AI inference supply-chain leaks2[1]. That study planted approximately two million prompts containing unique, trackable secrets into widely used AI routing services over five months from July through November 2026 and found that sensitive data leaked extensively across the AI inference supply chain[1].

Remediation and review

Together AI said its review has not identified evidence of additional prompt leakage beyond this incident. The company said it understands the data has now been deleted and has no reason to believe the data was leaked further. Together AI worked with an external forensic investigation firm and law enforcement. Its review is ongoing to determine whether sensitive personal information or other customer data was exposed. Together AI said it will contact affected customers directly as needed with findings relevant to them.

Zero data retention, abbreviated ZDR, is now the default setting for new accounts to limit prompt exposure risk. Together AI recommends that customers handling sensitive prompt data enable ZDR in their account settings. The company also updated development processes governing the use of AI coding agents, added credential checks on all pull requests, and moved to scoped or short-lived credentials and keyless access where supported.

ANALYSIS The incident illustrates a concrete risk in AI-assisted software development: an autonomous coding agent can inadvertently expose secrets in version-control commits faster than automated scanners can revoke them, creating an exploitation window even when detection works as designed. The Dragos canary-credential findings surfacing in the same breach connect the Together AI incident to a broader, documented pattern of prompt data leaking through AI inference infrastructure.

Was this useful?

The day’s vector, in your inbox.

Vector Wire Daily: one dispatch each publishing day.