Meta is rolling out new software that will log mouse movements, clicks, and keystrokes from its U.S. workforce. The system targets work‑related applications and injects periodic screenshots to give context for training. Company memos called the effort part of a Model Capability Initiative aimed at feeding real‑world data into future AI agents to improve their accuracy in daily tasks overall performance.

By collecting granular interaction data, Meta hopes to teach agents how to navigate menus, click buttons, and complete routine computer work that current models find tricky. CEO Andy Stone said the footage will not be used to evaluate employees, only to train systems. The initiative reflects a broader trend of companies harvesting internal usage data for AI across different domains.

The rollout applies only to U.S. staff, as European regulations would block similar monitoring. Meta has already faced EU scrutiny over forcing users to opt out of data use for AI training. If the new system succeeds, it could set a precedent for how tech firms balance employee privacy with the demand for richer training datasets that drive innovation today.