According to reports, Meta is exploring new sources of training data for its artificial intelligence systems, including company employees internally. Under the plan, Meta will record employee mouse movements, click behaviors, keyboard inputs, and other operational data within specific applications via internal tools to train more efficient and practical AI models. In response to media inquiries, Meta stated that building AI agents capable of assisting users in completing daily computer tasks requires models to learn based on real-world usage scenarios. Such data reflects how users navigate interfaces, operate menus, and execute workflows, thereby improving model performance in actual applications. The company emphasized that this tool operates exclusively within designated applications and incorporates protective mechanisms to prevent misuse of sensitive information, while ensuring related data remains unused for any other purposes. This initiative also underscores the ongoing expansion of data acquisition strategies across the AI industry. As high-quality training data becomes critical for enhancing model capabilities, tech companies continue seeking diverse channels to collect behavior patterns closely aligned with authentic user interactions. Previous reports indicated some enterprises have begun organizing records of internal communications—such as historical data from collaboration platforms and project management systems—as potential resources for AI model training. However, such practices have sparked discussions regarding privacy and data boundaries. Industry experts note that as internal corporate behavioral data increasingly contributes to AI training datasets, striking an appropriate balance between advancing model capabilities and safeguarding personal and organizational information will remain a persistent challenge for the sector.