Meta to start tracking employee keystrokes, mouse movements, and screen activity to train AI models

Meta is preparing to track how its employees use their computers, capturing keystrokes, mouse movements, and screen activity to train its AI models, according to an exclusive Reuters report.

Internal memos reviewed by the news outlet show the company plans to install software on work devices that records everyday interactions across apps and websites, turning routine tasks into training data for its next generation of AI systems.

The tool, called the Model Capability Initiative (MCI), runs across work apps and websites. It can even capture occasional snapshots of what’s on an employee’s screen. One memo, shared in an internal channel for Meta’s model-building team, framed the effort as a way to close a stubborn gap in AI performance. Current systems still struggle with everyday actions like navigating dropdown menus or using keyboard shortcuts.

“Meta is installing new tracking software on U.S.-based employees’ computers to capture mouse movements, clicks, and keystrokes for use in training its artificial intelligence models, part of a broad initiative to build AI agents that can perform work tasks autonomously,” Reuters reported, citing the company’s ‌internal memos to staffers.

“This is where all Meta employees can help our models get better simply by doing their daily work,” the memo read.

Meta Is Watching: New Tool Tracks Employee Keystrokes and Screens to Train AI

The move sits at the center of a broader push inside Meta to reshape how work gets done. The company has been reworking internal processes around AI, focusing on building agents that can perform tasks autonomously. In a separate memo, Chief Technology Officer Andrew Bosworth described the direction in plain terms: “The vision we are building towards is one where our agents primarily do the work and our role is to direct, review, and help them improve.” He added that agents should “automatically see where we felt the need to intervene so they can be better next time.”

To get there, Meta needs large volumes of real-world interaction data. That’s where MCI comes in. A company spokesperson, Andy Stone, confirmed that data collected through the system will be used to train the model. He said the information won’t be used for performance reviews and that safeguards are in place to protect sensitive content, though he did not detail what data would be excluded.

“If we’re building agents to help people complete everyday tasks using computers, our models need real examples of how people actually use them — things like mouse movements, clicking buttons, and navigating dropdown menus,” Stone said.

Inside Meta’s AI Training Push

The initiative reflects a wider shift across the tech industry. AI tools are getting better at handling complex workflows, from writing code to organizing data, with minimal human input. That progress is changing how companies think about staffing and operations. Meta has already signaled cuts, with plans to lay off about 10% of its global workforce starting May 20, and to make further reductions later in the year.

Other companies are moving in the same direction. Amazon has reduced its corporate workforce by tens of thousands over recent months. Fintech firm Block has also made deep cuts. Across the sector, executives are betting that AI can take on a growing share of knowledge work.

Inside Meta, that shift is already visible. Employees are being encouraged to rely on AI agents for coding and other tasks, even if it slows productivity at first. The company has introduced a broader job category called “AI builder,” blurring lines between traditional roles. It recently launched a new Applied AI engineering group focused on improving coding capabilities and building agents that can handle much of the work behind future products.

Strong engineers have already started moving into that team.

Rising Concerns Over Workplace Surveillance

Still, the plan raises new questions about workplace surveillance. Software that logs activity or captures screenshots has long been used to monitor productivity or detect misconduct. What’s different here is the purpose and the scale.

Ifeoma Ajunwa, who studies labor and technology, said the approach pushes white-collar monitoring into territory more commonly seen in gig work. Employees are now subject to a level of real-time tracking that mirrors what delivery drivers or warehouse workers have experienced for years. In the United States, there are few federal limits on this kind of monitoring, beyond basic notice requirements.

The picture looks different in Europe. Valerio De Stefano said similar practices would likely face legal challenges under stricter data protection rules. In some countries, including Italy, tracking employee activity electronically for productivity purposes is banned outright. German courts have set a high bar as well, allowing keystroke logging only under exceptional conditions.

Meta’s bet is that the payoff will be worth the tension. Better training data could help its AI systems handle everyday work on their own. That’s the future the company is building toward, one where software doesn’t just assist employees but increasingly replaces the need for them in certain tasks.

For now, that future starts with watching how people work.