Xiaomi AI Agent Lobster Makes a Quiet Appearance: Commitment Not to Use User Data for Training

Xiaomi Technology officially announced today that its new AI interaction exploration product, based on the Xiaomi MiMo large model, Xiaomi miclaw (internal code name "Crab") has started a small-scale closed test.

The product is positioned as a mobile AI Agent (intelligent entity), aiming to reconstruct the mobile human-computer interaction experience through four dimensions: system-level capabilities, personal context understanding, ecosystem interconnection, and self-evolution. Currently, the first batch of tests is only available to tech enthusiasts and geek users, prioritizing support for Xiaomi 17 series devices, with an invitation-only entry, and no large-scale public recruitment is currently being conducted.

Regarding the privacy and data compliance issues that have attracted attention, Xiaomi clearly responded that the product strictly follows the principle of data minimization throughout the entire R&D lifecycle and commits to "never using user data to train the AI system."

According to official disclosure, all model training of Xiaomi miclaw relies on legally published datasets or authorized data that have passed compliance review. In terms of interaction execution, real-time user commands are only used for task distribution and execution, and do not enter the training resource pool. In addition, core private data is processed primarily on the phone's local side using "end-cloud privacy computing" technology, blocking sensitive information from being uploaded to the cloud at both physical and algorithmic levels.

Currently, Xiaomi miclaw is still in the technical exploration phase. Xiaomi frankly acknowledges that there is room for improvement in power consumption performance, success rate in complex scenarios, and system stability.

The launch of this closed test marks that smartphone manufacturers are accelerating their transition from "integrating AI functions" to "native AI assistants." Xiaomi attempts to enhance the autonomous execution capability of the Agent while establishing a stricter industry data security benchmark by combining end-side models with privacy computing. This is not only a deep preview of its future AI phone form but also will drive smartphone end-side large models from theory into practical application in complex scenarios.