You send Atlas Cloud prompts, code and files that matter to your work. You should know how we handle them, and what we do when we learn of a security issue at a provider upstream of us.
How we handle your data
- We don't sell your data or train our models on it.
- You choose how long we keep your request history and generated files, in your account settings.
- Third-party models are processed under each provider's terms.
- We're reviewing our other upstream providers for similar exposure.
An issue with MiniMax's official API
Security researcher HD Moore published findings showing that a third party opened private links placed in test prompts sent to MiniMax's official API, including prompts sent to MiniMax M3 and M2.7 through Atlas Cloud in Q3 and early Q4 2026. So the prompts and outputs of other requests to these models may also have been accessed. The access happened after the requests reached MiniMax, not on Atlas Cloud's infrastructure. Atlas Cloud did not store the prompts or outputs of those requests.
We stopped sending requests to MiniMax's official API, and no Atlas Cloud request goes there until we're confident in its security. MiniMax models stay available only where we host them ourselves or the hosting provider confirms requests don't reach MiniMax's official API. We're emailing every account that used these models.
If you used MiniMax M3, M2.7 or M3.1 Flash on Atlas Cloud, we can't tell whether data from your requests was misused, so treat any secret you sent to these models as exposed. Check your prompts and outputs for API keys, passwords, tokens and private URLs, and revoke any exposed credentials, tokens or links. The email we sent has the full steps.
Contact
Questions about your account: [email protected]. Privacy questions: [email protected]. We'll update this post if we learn anything that changes this guidance.







