Reuters recently revealed that Meta is planning to train its next generation of artificial intelligence models by collecting real-time work data from its employees, aiming to build AI systems that can perform tasks and respond to queries more efficiently. According to reports, Meta will launch a specialized internal tool to track and record employees' mouse movements, keyboard inputs, button clicks, and menu navigation within specific applications. This move marks a shift for tech giants facing pressure from external data shortages, as they begin to explore internal "clickstream" data to enhance the understanding and adaptability of AI agents in daily office tasks.
A Meta spokesperson confirmed the plan in a statement, emphasizing that to build intelligent agents capable of truly assisting users in operating computers, the model must learn real examples of human computer usage. Meta stated that it has already taken security measures to protect sensitive content and promised that the data would only be used for model training and not for any other purposes. However, this strategy has also sparked widespread discussions in the industry about the boundaries of privacy.
Currently, the artificial intelligence industry is at a critical stage of model iteration, with high-quality training data becoming a core competitive asset. Last week, reports also indicated that internal communications records, such as Slack archives and Jira tickets from some established startups, are being converted into AI training material. Meta's latest move further confirms the industry trend: internal corporate communication and operational behaviors are increasingly becoming key "fuel" in the new supply chain. Although this shift from publicly available internet data to closed, high-frequency human behavior data may lead to breakthroughs in AI performance in professional productivity fields, it also signals that the balance between workplace privacy and technological advancement is becoming increasingly complex.
