Today's AI/ML headlines are brought to you by ThreatPerspective

Digital Event Horizon

Prompt Leakage Incident Raises Concerns over Security and Data Protection in AI Inference Services


Together AI has announced a security incident involving prompt leakage for requests in June and July of 2026, highlighting the need for robust security measures and data protection protocols in AI inference services. The incident has led to strengthened controls and awareness among customers to protect themselves against similar incidents.

  • A recent security incident involving Together AI's serverless inference service raised concerns over prompt leakage and data protection.
  • An unauthorized party downloaded stored prompt samples from a private Hugging Face repository due to an exposed credential during routine software testing.
  • Together AI strengthened controls around development workflows, credentials, and customer data to prevent similar incidents.
  • Customers handling sensitive prompt data should enable zero data retention in their account settings to protect themselves.
  • The incident highlights the need for robust security measures and data protection protocols in AI inference services.



  • A recent security incident involving Together AI's serverless inference service has raised concerns over the potential for prompt leakage and data protection in the use of AI models for inference. In June and July of 2026, it was discovered that an unauthorized party had downloaded stored prompt samples from a private Hugging Face repository, which had been compromised due to a exposed credential. The incident highlights the need for robust security measures and data protection protocols to prevent similar incidents in the future.

    The incident occurred as part of routine software testing, where prompt samples authorized for that use were collected and stored in a private repository. However, during development of the public GitHub repository, an AI coding agent committed a credential that provided access to the private repository, allowing the unauthorized party to download the stored prompt samples.

    The exposed credential was later discovered and reverted, but not before the unauthorized party had obtained access to the prompt samples. The party subsequently searched those samples for credentials to other services and tested whether they were valid, highlighting the potential risks of prompt leakage.

    Together AI has since strengthened controls around development workflows, credentials, and customer data to prevent similar incidents in the future. These changes include using scoped or short-lived credentials and keyless access where supported, checking all pull requests for credentials, making zero data retention the default setting for new accounts, and updating development processes governing the use of AI coding agents.

    Customers handling sensitive prompt data should be aware of the potential risks and take steps to protect themselves, such as enabling zero data retention in their account settings. Together AI has notified affected customers directly and will contact them with findings relevant to their organization.

    The incident serves as a reminder of the importance of robust security measures and data protection protocols in AI inference services. It highlights the need for developers and users to be vigilant and take proactive steps to protect themselves against similar incidents.



    Related Information:
  • https://www.digitaleventhorizon.com/articles/Prompt-Leakage-Incident-Raises-Concerns-over-Security-and-Data-Protection-in-AI-Inference-Services-deh.shtml

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


  • Published: Sat Nov 21 17:53:11 2026 by llama3.2 3B Q4_K_M











    © Digital Event Horizon . All rights reserved.

    Privacy | Terms of Use | Contact Us