Meta Begins Collecting Employees' Mouse Movements and Keystrokes to Train AI

Meta has installed new tracking software on US-based employees' work computers to collect mouse movements, clicks, and keystrokes, and uses this data to train AI models.

The tool operates within work-related apps and websites, and also intermittently saves partial snapshots of screen content. The goal is to help the AI better learn how people actually operate computers, so it can more accurately reproduce interactions such as dropdown selections or keyboard shortcuts.

Meta explained that this data is not used for performance evaluations, and is not used for any purpose other than model training. However, the company did not disclose specific criteria for what sensitive content is excluded.

This move is part of Meta's broader push to shift internal workflows toward an AI-agent-centric approach. In an internal memo, the company laid out a direction where agents primarily perform tasks while humans focus on instructing, reviewing, and improving them.

Key points:

• Introduction of keylogging-level monitoring for US employees
• Mouse/click/keystroke input + screen snapshots used as AI training data
• Purpose is to improve the performance of AI agents that perform computer tasks
• Meta states the data is not used for performance evaluations