Meta Deploys Keystroke-Tracking Software on Employee Computers to Train AI Agents

Meta rolls out employee surveillance tool for AI training

Meta has begun installing new software on the work computers of its US-based employees that tracks mouse movements, clicks, and keystrokes. According to Reuters, which obtained internal documents, the collected data is being used to train AI models as part of a broader effort to build autonomous AI agents capable of performing work tasks independently.

The tool, called Model Capability Initiative (MCI), operates within work-related applications and websites. It is also configured to periodically capture screenshots of employees' screens.

Why this matters

Meta's initiative illustrates how major tech companies are turning everyday employee activity into training fuel for AI systems. The company is effectively building a feedback loop where workers train the very agents that could eventually handle parts of their jobs. This sets a significant precedent for the industry — particularly amid widespread layoffs across the tech sector.

How MCI works and its stated purpose

The initiative aims to improve AI models in areas where they still underperform. Examples include navigating dropdown menus and using keyboard shortcuts.

Internal documents state that every Meta employee can help the company's models improve simply by going about their daily work.

Meta CTO Andrew Bosworth wrote in a separate memo that the company is ramping up internal data collection under its AI for Work initiative. He stated that the goal is for agents to do the work while employees guide them, review their output, and help them get better.

Meta spokesperson Andy Stone said that data collected through MCI will not be used for performance evaluations or any purpose other than model training. The company has also put safeguards in place to protect confidential content.

Automation amid mass layoffs

AI tools are enabling companies to automate tasks and cut staffing costs — a trend that several major firms are actively pursuing:

• Meta plans to lay off 10% of its workforce starting May 20, with additional large-scale cuts being considered by year-end.
• Amazon has terminated contracts with 30,000 workers over the past several months.
• Block cut nearly half its staff in February.

At the same time, Meta is encouraging the use of AI agents for coding and other tasks, even if it slows things down in the short term. Last month, the company formed an Applied AI engineering team tasked with enhancing neural network coding capabilities and building agents. These agents are intended to be used for developing, testing, and shipping future products and infrastructure.

In early April, Meta began transferring its strongest programmers into the Applied AI division.

Legal implications

Ifeoma Ajunwa, a professor of law at Yale University, noted that recording computer activity and taking screenshots have historically been used by companies to detect employee misconduct. However, the shift to keystroke logging takes surveillance to an entirely new level — office workers now face the kind of real-time monitoring previously experienced only by delivery drivers and gig economy workers.

According to Ajunwa, there are no federal-level restrictions on worker surveillance in the United States. In contrast, Valerio De Stefano, a professor of law at York University in Toronto, emphasized that European legislation would likely prohibit this type of monitoring.

Earlier in April, Meta unveiled its Muse Spark AI model, developed by the newly formed Meta Superintelligence Labs research team.