On September 21, the Zhipu MaaS platform announced that it will soon launch a "data content not stored" feature, offering enterprises and developers stricter data privacy protection. The core logic of this mechanism is: input and output data generated when users call the model will no longer be stored statically on the platform, and will only be used to complete the current model call, being discarded afterward.
According to the official statement, any user can submit an application through the MaaS console. The specific implementation time and applicable scope will be determined by the platform. This means that for enterprise customers with data compliance requirements, it adds an optional "data insurance" layer.
However, this mechanism is not an absolute "no storage at all" approach. Zhipu clearly stated that functions such as Batch API and File API, which require persistent storage of tasks or files on the platform side, are not covered by this mechanism. In addition, the platform may still retain relevant data for 30 days or more as required by laws and regulations, or for checking suspected violations or abuse behaviors.
At the current stage where large AI models are accelerating their deployment, data privacy has become a key consideration for enterprises when making selection decisions. The "no storage" mechanism launched by Zhipu essentially offers a compromise that leans more toward privacy protection between model calling efficiency and user data sovereignty — data is burned after use, but the bottom line for compliance and risk control is still maintained.
