Meta Platforms is deploying software on all [1] U.S.-based employee computers to record mouse movements, keystrokes, and screenshots.
This move signals a shift in how artificial intelligence is developed, moving from static datasets to the active surveillance of professional workflows. By capturing real-time human behavior, Meta aims to create AI agents that can mimic complex workplace tasks with higher precision.
The company is routing this captured data into a program called the Model Capability Initiative [1]. The software monitors a wide array of work activity, including clicks and screen captures, to gather large-scale, real-world data [1]. Meta said this information is necessary to train and improve its artificial intelligence agents [1].
The tracking program was announced in March 2024 [1]. The rollout ensures that the company can observe how employees navigate software and solve problems in a live environment. This level of granular monitoring allows the AI to learn the specific sequences of actions humans take to complete professional assignments.
While the company frames the initiative as a technical necessity for AI evolution, the deployment of such invasive tools on all [1] company computers in the U.S. raises questions about the boundary between productivity and privacy. The software effectively turns the daily operations of the workforce into a training set for the very technology that may eventually automate those roles.
“Meta is rolling out software that records mouse movements, keystrokes, screenshots, and other work activity.”
The integration of employee surveillance into AI training represents a move toward 'imitation learning' at scale. By recording the exact digital footprints of its workforce, Meta is attempting to bridge the gap between general AI and specialized professional agents. This creates a feedback loop where human labor is used to build the tools that may eventually replace or augment those same human roles.
