跳到正文
Together AI 研究与产品博客·· 1 天前精选AI 评分74

Together AI 通报 2026 年 6-7 月 serverless 推理服务提示词泄露事件

Security event: Prompt leakage for requests in June and July of 2026

AI 导读

Together AI 通报一起安全事件:因 AI 编码 agent 在公开 GitHub 仓库 xorl 中提交了私有 Hugging Face 仓库凭证,未授权方下载了 2026 年 6-7 月提交给 MiniMax M3、GLM-5.2、Kimi K2.6 等 serverless 推理服务的存储提示词样本,并搜索其中其他服务凭证。

推荐理由

原文披露了提示词泄露的范围、成因与已落实的整改措施,用户可据此评估自己是否受影响及是否开启 ZDR。

正文 · 原文

We are notifying customers about a security incident involving a subset of prompts submitted to Together AI’s serverless inference service in June and July 2026. An unauthorized party downloaded stored prompt samples after a credential for a private repository was exposed in a public code repository.

We apologize for this incident. We are reviewing the affected data to determine what information was exposed and which customers require direct notification.

What happened

As part of routine software testing, we collected prompt samples authorized for that use and stored them in a private Hugging Face repository. The samples came from requests to the MiniMax M3, GLM-5.2, and Kimi K2.6 models, primarily on June 24, July 4, July 8, and July 14, 2026.

During development of our public GitHub repository, xorl, an AI coding agent committed a credential that provided access to the private Hugging Face repository. Our automated scanning detected the exposed credential, and we reverted the commit.

During the exposure window, an unauthorized party obtained the credential and used it to download the stored prompt samples. The party subsequently searched those samples for credentials to other services and tested whether those credentials were valid. Five of the credentials tested were canary credentials associated with a Dragos security experiment.

We have no reason to believe the unauthorized party has leaked this data further and understand that it has now been deleted.

Investigation and scope

We have worked with an external forensic investigation firm and law enforcement to investigate the incident.

Based on our review, we have not identified evidence of additional prompt leakage beyond this incident. Our review of the affected prompt contents is ongoing to determine whether sensitive personal information or other customer data was exposed. We will contact affected customers directly as needed with the findings relevant to them.

Changes we have made

Following the incident, we strengthened controls around development workflows, credentials, and customer data. These changes include:

  1. Using scoped or short-lived credentials, and keyless access where supported, to reduce the potential impact of credential exposure.
  2. Checking all pull requests for credentials.
  3. Making zero data retention (ZDR) the default setting for new accounts.
  4. Updating our development processes governing the use of AI coding agents.

What customers can do

Customers handling sensitive prompt data should enable ZDR in their account settings.

If you have questions about this incident or its potential impact on your organization, contact security@together.ai.

‍

Vipul Ved Prakash

CEO and Co-founder, Together AI

来源:Together AI 研究与产品博客 · together.ai