How did Meta train AI using employee input?

Meta starts capturing employee computer activity for AI training

Meta is installing software on some employees’ computers to collect high-granularity usage signals—specifically mouse movements, clicks, keystrokes, and, in some cases, screenshots of what’s on the screen. The stated goal is to generate better training data for Meta’s AI systems, especially as the company pushes toward AI agents that can navigate and act within software tools.

What Meta says it’s collecting

The activity monitoring is framed as a way to understand how people interact with applications in real workflows. That means the logs aren’t just about model prompts; they’re about observing the steps employees take while using tools—where they click, what they type, and what information appears on their screen as they work.

Why it matters now

This approach raises immediate privacy and labor implications because it turns everyday office behavior into data that can be repurposed for machine-learning training. It also highlights how “agentic AI” depends on operational know-how: building systems that can reliably operate in real apps requires examples of how humans actually perform tasks.

What’s next for users and employees

Meta’s move is likely to intensify questions about: - Workplace surveillance boundaries (what’s collected, how long it’s stored, and who can access it) - Consent and transparency for employees participating in training data collection - Data governance for screen and keystroke signals, which are typically far more sensitive than normal application telemetry

As more AI features transition from chat to action, companies are increasingly looking to “behavioral datasets” collected from real-world interactions. Meta’s announcement is a clear example of that shift—and it may become a template for other firms if not checked by employee pushback and regulation.