🤖 AI Summary
Together.ai has informed customers of a security breach in which a subset of prompts submitted to its inference service was accessed by an unauthorized party due to the exposure of a credential related to a private repository. This incident, which occurred during routine software testing between June and July 2026, involved prompt samples collected from interactions with several AI models, including MiniMax M3 and GLM-5.2. Although the credential was quickly retracted upon discovery, the unauthorized access allowed the hacker to download stored prompts and test them for additional credentials.
This breach raises significant concerns within the AI/ML community, especially regarding data security and the handling of sensitive information. Together.ai is now conducting a thorough investigation to determine what information may have been exposed and is collaborating with a forensic firm and law enforcement. In response to the incident, the company has implemented enhanced security measures, such as using short-lived credentials and making zero data retention the default setting for new accounts, to mitigate future risks. Customers dealing with sensitive data are advised to enable zero data retention in their settings and can reach out to Together.ai for further inquiries about the incident.
Loading comments...
login to comment
loading comments...
no comments yet