Why did Meta start recording employee keystrokes?

Meta plans to turn employee activity into AI training data

Meta announced it will begin collecting detailed on-device activity from its employees—specifically mouse movements, clicks, keystrokes, and occasional screen snapshots—to generate training data for future AI agents.

The core change is that the material being gathered is not just traditional log data or clicks on software interfaces. Instead, Meta wants “how people use tools” signals: how staff interact with apps moment-to-moment, and how those interactions correlate with the work they are trying to do.

What Meta is collecting

• Mouse movements and clicks
• Keystrokes
• Occasional screen snapshots
• Data tied to work-related app usage

The company’s stated goal is higher-quality training material for AI agents that can operate with more context and fewer mistakes. In practice, that means the training set is intended to include richer examples of user intent and workflow steps, rather than relying solely on text-only prompts or aggregated telemetry.

Why it matters

This development is significant because it raises the stakes around workplace data collection and model training. It also reflects a broader industry shift: as agentic systems become more capable, companies are looking for training signals that capture the entire interaction loop—not just the final request.

Across the tech sector, these efforts can influence how regulators and employees evaluate consent, privacy expectations, and security safeguards. Even if Meta is not changing public-facing product behavior, the data collection approach can set a precedent for how other AI builders think about “in-the-loop” training sources.

It’s still unclear what protections Meta will use to limit access to sensitive employee data, how long it will retain the material, or how broadly it will deploy the system beyond initial pilots.