AI Agent News/stories/together-ai-discloses-prompt-leakage-from-june-and-july-2026-inference-requests.md

InfrastructureBreaking

Together AI discloses prompt leakage from June and July 2026 inference requests

Together AI is notifying customers that an unauthorized party downloaded stored prompt samples from its serverless inference service in June and July 2026. The access began when an AI coding agent committed a credential to a public GitHub repository, exposing a private Hugging Face repository.

Source: www.together.ai/blog/security-event-prompt-leakage-for-requests-in-june-and-july-of-2026

Together AI is notifying customers about a security incident involving a subset of prompts sent to its serverless inference service in June and July 2026. The stored prompt samples came mainly from requests to MiniMax M3, GLM-5.2 and Kimi K2.6 on June 24, July 4, July 8 and July 14. The company says it had collected those samples for routine software testing and kept them in a private Hugging Face repository.

An AI coding agent committed a credential during development of the public xorl GitHub repository, and that credential opened the private repository. Automated scanning detected the exposure and the commit was reverted, but an unauthorized party had already obtained the credential and downloaded the samples. The party searched them for credentials to other services and tested whether they worked, and five of the tested credentials were canary credentials tied to a Dragos security experiment.

Together AI says it has no reason to believe the data has spread further and understands it has been deleted. An external forensic firm and law enforcement are involved, and the company has found no evidence of other prompt leakage. A review of the affected prompt contents is still underway to determine whether personal information or other customer data was exposed, and affected customers will be contacted directly.

Why it matters if you build agents

Developers who sent prompts to Together AI's serverless inference in that period should assume those prompts may have been accessed and check whether they contained sensitive data. Teams handling sensitive prompt data can enable zero data retention in account settings, which is now the default for new accounts. Agent builders should also treat credentials handled by AI coding agents as a leak risk and prefer scoped or short-lived credentials.

Get stories like this every weekday

One short email with the day's agent tooling news.

Subscribe