← In the News

Together AI says an AI coding agent committed the credential behind its prompt leak

Security event: Prompt leakage for requests in June and July of 2026 · Vipul Ved Prakash, CEO and co-founder, Together AI · Together AI blog, November 21, 2026

Machine-readable Download Markdown

Together stored prompt samples for software testing in a private Hugging Face repository. The samples came from requests to MiniMax M3, GLM-5.2 and Kimi K2.6, mainly on June 24, July 4, July 8 and July 14, 2026. The post says that while the company was developing its public GitHub repository, xorl, "an AI coding agent committed a credential that provided access to the private Hugging Face repository." Automated scanning caught the credential and the commit was reverted, but during the exposure window an unauthorized party took it, downloaded the samples, and searched them for credentials to other services. Five of the credentials it tested were canaries from the Dragos experiment.

Together lists four changes: scoped or short-lived credentials and keyless access where supported, credential checks on every pull request, zero data retention (ZDR) as the default for new accounts, and updated development processes for AI coding agents. The post does not say which agent was involved or what the new agent processes are. Its review of what the prompts contained is still in progress.

Why it matters: This is a provider's own post-mortem in which the leaked credential was committed by an agent. Scanning caught the secret after it landed, and that was not fast enough. If your agents can commit to a public repository, put the secret check in front of the commit, and keep the credentials those agents can read short-lived and narrowly scoped.