Meta Is Recording Employee Screens, Clicks, and Keystrokes to Train AI That Could Replace Them

Meta just told its US employees their every click now feeds an AI model. The company confirmed it installed tracking software on worker computers to capture mouse movements, keystrokes, and random screenshots. The program runs inside a division called Meta Superintelligence Labs and targets specific work apps and websites. Meta calls this the Model Capability Initiative. Its stated goal is to build AI agents that perform white-collar tasks without human input.

Key Takeaways

• The tracking tool is called the Model Capability Initiative
• It runs on designated work apps and captures real behavioral data
• The data trains AI agents to replicate basic computer-use tasks
• US federal law places no ceiling on this kind of employee surveillance
• Meta plans to cut 10% of its global workforce in the months ahead

What Meta Employee Surveillance Looks Like in Practice

The Model Capability Initiative records employee screens as they work. It captures mouse movements, button clicks, and keyboard inputs across a set list of work-related apps and websites.

An internal memo sent to staff in a Meta Superintelligence Labs channel, obtained by Reuters, framed the effort as a way for employees to improve company models in areas where AI struggles to replicate basic computer-use behaviors. The memo told workers they could help by just doing their regular jobs.

The tool also randomly captures images of workers’ screens at timed intervals, giving the model visual context alongside behavioral input data.

What Apps and Data Are in Scope

The tool does not run across every app on a work computer. It targets a designated list of work apps and websites, and Meta has stated that safeguards are in place to protect sensitive content.

A Meta spokesperson confirmed the scope to TechCrunch: the captures include things like mouse movements, clicking buttons, and navigating dropdown menus. The company says the data goes toward one purpose only.

Why Meta Needs Real Keystrokes to Train AI Agents

AI models cannot learn computer navigation from text alone. They need labeled examples of real people clicking through real interfaces to replicate that behavior.

The memo framed the effort as a solution to a gap in AI capability: models struggle to emulate basic computer-use behaviors, such as navigating dropdown menus and using keyboard shortcuts. Synthetic or simulated data does not capture the natural variation in how different users complete the same task.

The broader goal is to build AI agents capable of performing white-collar tasks on their own, the same category of software Meta races to ship amid competition from OpenAI and Anthropic.

Are you already using AI tools at work without knowing how your inputs get logged or used?

The Scale AI Connection

Meta acquired a 49% stake in data-labeling firm Scale AI for more than $14 billion. Scale’s former CEO, Alexandr Wang, now leads Meta Superintelligence Labs. That appointment places one of the most prominent figures in commercial data labeling directly over the team running this surveillance program.

Workplace behavioral data is now the most scarce input for the next generation of AI agents. Separately, it was reported that old startups are being scavenged for their corporate communications, including Slack archives and Jira tickets, and converted into AI training data. Meta’s approach cuts out the middleman entirely.

The Legal Position on Meta Employee Surveillance in the US

Yale University law professor Ifeoma Ajunwa told Reuters that there is no limit on worker surveillance at the federal level in the United States. Employers can legally track keystrokes and mouse movements on company hardware.

Tracking of this kind would likely violate European law. Meta’s program currently targets US-based employees only. That geographic boundary keeps the program legally compliant for now while leaving workers with little recourse.

This gap between US and European worker protections is not new. What makes this case different is the stated endpoint: the data directly trains a system positioned to automate the same jobs producing it.

Workforce Cuts Running Alongside the Program

Beyond tracking employee activity, Meta also plans to slash ten percent of its workforce across the globe starting next month, with additional cuts planned later in the year.

Meta has reportedly been building AI agents meant to work alongside employees, including one designed for CEO Mark Zuckerberg, and has created a Zuckerberg chatbot for staff to communicate with. Workforce reduction and AI agent development are moving on the same timeline.

What Other Companies Are Doing With Employee Data

Meta did not start this trend. In January, OpenAI was reported to be asking third-party contractors, via training data firm Handshake AI, to upload samples of real work products from previous jobs, including actual PowerPoints and spreadsheets, with instructions to scrub confidential material before submission.

The race to collect authentic workplace behavioral data is accelerating across every major AI lab. Each company faces the same problem: AI agents trained on public web data cannot replicate the specific, repetitive, interface-level behaviors that define most knowledge work.

Does knowing this change how you approach what you type and click on your work computer?

For more on how Meta is reshaping its product ecosystem alongside these AI moves, read Cloudorian’s coverage of the new WhatsApp Plus subscription, a separate Meta shift that changes what users exchange for access to features.

What to Do If You Work in Tech Right Now

Start with your employment contract. Look for any clause covering computer monitoring, data collection, or AI training participation.

Ask your HR or legal team a direct question: does your company monitor computer activity to train AI systems? Get the answer in writing.

1. Pull up your employment agreement and search for terms like “monitoring,” “data collection,” and “AI training.”
2. Request a written policy statement from HR on computer surveillance scope and purpose.
3. Check whether your company has disclosed any AI training programs tied to internal tools you use daily.
4. If you work in the EU, ask specifically whether your employer’s practices comply with your regional data protection rights.

Meta’s decision to log worker behavior as training data sets a precedent. Other companies watching the rollout will benchmark against it. Your next step is knowing where your employer stands before the same program lands on your machine.