Together AI disclosed that an unauthorized party downloaded a subset of prompt samples submitted to its serverless inference service during June and July 2026. The samples, collected for routine software testing with authorization, came from requests to MiniMax M3, GLM-5.2, and Kimi K2.6, mainly on June 24 and July 4, 8, and 14. An AI coding agent committed a credential for a private Hugging Face repository to Together’s public GitHub repository. Although automated scanning detected the exposure and the commit was reverted, an attacker obtained the credential during the exposure window and used it to download the samples. The attacker searched them for credentials to other services and tested some; five tested credentials were canaries used in a Dragos security experiment.
Together has worked with an external forensic firm and law enforcement. It reports finding no evidence of prompt leakage beyond this incident and says it has no reason to believe the downloaded data was shared further and understands it has been deleted. The review of prompt contents remains underway to establish whether personal or other customer information was exposed; customers affected will be contacted directly. In response, Together says it has tightened development and data-handling controls, including adopting scoped or short-lived credentials and keyless access where possible, scanning pull requests for secrets, and changing processes for AI coding agents. Zero data retention is now the default for new accounts. Customers with sensitive prompts are advised to enable it in their settings and can contact Together’s security team with questions.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.