The search for high-quality AI training data has moved from the public internet into the private cubicle. Meta is now harvesting the digital footprints of its own employees—specifically their keystrokes and mouse movements—to build more capable artificial intelligence.
The Anatomy of a Click
As first reported by Reuters, Meta is deploying an internal tool to track how staff interact with various applications. This isn’t just about what employees are typing; it’s about the mechanics of how they work. The data points being collected include:
- Navigation patterns: How users move through dropdown menus and buttons.
- Input dynamics: The specific timing and sequence of keystrokes.
- Cursor behavior: The path and speed of mouse movements across the screen.
Meta’s objective is to develop AI agents that can automate “everyday tasks.” To do this effectively, the models require real-world examples of how humans navigate software interfaces. By watching its own experts work, the company hopes to teach its models to mimic human fluidity and efficiency.
Privacy in the Age of Scavenged Data
While Meta told TechCrunch that it has implemented safeguards to protect sensitive content and ensure the data is used solely for training, the move signals a shift in corporate privacy. The tech industry is increasingly viewing internal communications and behavioral data as “AI fuel.”
This trend extends beyond Meta. Recent reports indicate that even defunct startups are having their Slack archives and Jira tickets scavenged for training purposes. As the demand for data grows, the line between workplace activity and training material continues to blur, turning every click into a potential lesson for the next generation of AI.
