Technology

Together AI says coding agent leaked user prompts

Together AI said an AI coding agent exposed a credential, leaking serverless prompts from June and July 2026.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

2 min read

Together AI serverless inference security disclosure
Credit: Together AI

Key facts

  • MiniMax M3, GLM-5.2 and Kimi K2.6 requests were affected
  • Five tested credentials were Dragos canaries, per Together AI
  • ZDR is now default for new Together AI accounts

Together AI disclosed prompt leakage affecting serverless inference requests from June and July 2026. The samples came from requests to MiniMax M3, GLM-5.2 and Kimi K2.6, primarily on June 24, July 4, July 8 and , the company said. Access was remote. The intruder reused a Hugging Face credential an AI coding agent had committed to the public xorl repository before Together AI reverted it.

If you run sensitive prompts through shared inference, turn on zero data retention now.

Together AI said it had collected the samples for routine testing and stored them in a private Hugging Face repository. Its automated scanning caught the exposed credential and reverted the commit, but not before download, the company said. The party then searched the prompts for other service credentials and tested them. Five hits were canary credentials from a Dragos security experiment, the company said. Ars Technica reported the broader canary operation planted 2 million canaries and saw secrets sent to two hacked providers used together almost immediately from the same rented server.

That last detail hurts. It suggests at least two providers were hit by the same adversary, and stolen keys get tried fast. Together AI said it has no reason to believe the party leaked the data further and understands it has now been deleted. That claim rests on investigation, not on proof of containment, since downloaded prompts cannot be pulled back.

The company said it worked with an external forensic firm and law enforcement and found no evidence of additional leakage beyond this incident. Its review of prompt contents is still underway to decide who gets direct notice, the company said. CEO and co-founder Vipul Ved Prakash signed the disclosure. Questions go to [email protected], the company said.

For teams, the workflow change is explicit. Together AI said it moved to scoped or short-lived credentials and keyless access where supported, added credential checks on all pull requests, made zero data retention the default for new accounts, and tightened development rules for AI coding agents. If your account is older, you do not get the new default. Open account settings, enable ZDR, and rotate any secret sent to those three models on those four dates.

Mentioned in this story

Sources

TopicsTogether AIMiniMax M3GLM-5.2Kimi K2.6Hugging FaceDragosVipul Ved Prakash

StartupHub.ai StaffDaniel SingerWritten by StartupHub.ai Staff, Editorial teamReviewed by Daniel Singer, EditorAbout

StartupHub.ai Staff, Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer, Editor

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

More from Daniel Singer

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.