Meta Records Employee Keystrokes to Train AI That Could Replace Them

A computer keyboard and mouse on an office desk with a monitor in the background — Photo via Unsplash

TL;DR: Meta is installing tracking software called “Model Capability Initiative” on U.S. employees’ work computers. It captures mouse movements, keystrokes, clicks, and periodic screenshots from work applications including Gmail, GChat, Visual Studio Code, and internal tools. The data feeds Meta’s push to build AI agents that can perform white-collar tasks autonomously. Meta says the data won’t be used for performance reviews. Employees are skeptical. The company is simultaneously planning 20% workforce cuts starting in May 2026. There is no federal law stopping any of this.

Your Employer Watches You Type. Meta Made It Official.

On April 21, 2026, Reuters reported that Meta distributed an internal memo to the Meta Superintelligence Labs team announcing a new program called the “Model Capability Initiative” ^[1]. The initiative installs monitoring software on all U.S.-based employees’ computers. Here’s what it captures:

Every mouse movement: where you move, where you click, how fast
Every keystroke: what you type, when you type it
Periodic screenshots: snapshots of what’s on your screen for “context”
App interactions: how you navigate dropdown menus, use keyboard shortcuts, switch between windows

The tracked applications include Gmail, GChat, Visual Studio Code, and Meta’s internal platform Metamate ^[2]. The memo framed it as a way for “rank-and-file employees to improve company models in areas where they struggle to emulate basic computer-use behaviors” ^[1].

Translation: Meta is turning its workforce into a training dataset.

Building the Agents That Replace the Workers Training Them

Meta spokesperson Andy Stone told Fortune the data would “not be used for performance assessments or any other purpose besides model training,” and that “safeguards were in place to protect sensitive content” ^[1].

Meta’s CTO Andrew Bosworth was more blunt about the end goal, describing a future “where our agents primarily do the work and our role is to direct, review and help them improve” ^[2].

Connect the dots. Meta is:

Recording how employees do their jobs
Feeding that data into AI agents designed to do those same jobs
Planning to cut 20% of its workforce starting May 2026 ^[3]
Spending up to $135 billion on AI infrastructure this year ^[1]

The Meta Superintelligence Labs team behind the initiative is led by Alexandr Wang, formerly CEO of Scale AI. Meta acquired a 49% stake in Scale AI for over $14 billion ^[1]. Scale AI’s entire business model is turning human activity into training data. Now that’s Meta’s internal strategy too.

The Company Built on Surveillance Gets Surveilled

Here’s where it gets rich. Meta, the company that built a $1.5 trillion empire by tracking what 3.98 billion people click, type, watch, buy, and say, is now doing the same thing to its own employees ^[2].

The Register called it “magnificent irony” ^[2]. It’s the same company that:

Paid a $375 million verdict in New Mexico for failing to protect children
Faces an ACLU coalition of 75+ organizations demanding answers about facial recognition plans for Ray-Ban smart glasses
Has racked up multiple GDPR violations and FTC consent decrees for privacy failures
Was caught tracking employee badge swipes to enforce return-to-office mandates

Now Meta’s own employees get to experience what users have felt for 20 years: the uneasy realization that everything you do is being watched, recorded, and fed into a system you don’t control.

No Federal Law Stops This

Yale law professor Ifeoma Ajunwa told reporters there is “no limit on worker surveillance” under U.S. federal law ^[3]. Employers can legally monitor everything on company-owned devices. Keystrokes, screenshots, location data, browsing history: all fair game.

The Electronic Communications Privacy Act of 1986 includes a “business purpose” exception so wide you could drive a surveillance truck through it. If the employer owns the device and claims a business reason, monitoring is legal. Period.

European employees likely get more protection. The EU’s GDPR and national labor laws impose stricter consent requirements and proportionality tests for workplace monitoring. Meta said the initiative targets “U.S.-based employees” (geography chosen by legal, not engineering) ^[2] ^[3].

Meta Joins the Bossware Boom

Meta isn’t inventing employee surveillance. It’s just joining an industry that exploded after the pandemic forced offices remote. The “bossware” market (software that monitors employee activity on work devices) has grown into a multi-billion-dollar industry.

But Meta’s version is different in one critical way: the data doesn’t just monitor productivity. It feeds directly into AI systems designed to automate the monitored workers out of their jobs. That’s a new category of workplace surveillance: one where you’re not just watched, you’re training your own replacement.

Other tech giants are building similar AI agent capabilities. OpenAI has “Operator,” Anthropic and Microsoft have their own agent frameworks. The race to build AI that can navigate computers like humans requires data about how humans navigate computers. Employees are the most convenient source.

If Your Employer Does This: Know Your Rights

Ask what’s being collected. Your employer may not be required to tell you, but ask anyway. Get it in writing
Check your state laws. Connecticut, Delaware, and New York require employers to notify workers of electronic monitoring. California’s CCPA has workplace data provisions taking effect
Keep personal activity off work devices. If your employer monitors keystrokes, every personal message, password, and search query on that device is captured
Use your phone for personal business. Not company Wi-Fi: your personal phone on cellular data
Document the monitoring. If you’re eventually laid off and replaced by AI trained on your work, that documentation matters for potential legal action
Support the STOP Spying Bosses Act and similar legislation that would restrict workplace surveillance

The Quiet Part Out Loud

Meta could have collected this training data from consenting external testers. It could have synthesized computer-use data. Instead, it chose to capture the real work activity of tens of thousands of employees, many of whom are about to lose their jobs to the AI those keystrokes train.

When Andy Stone says the data “won’t be used for performance assessments,” remember: Meta said Facebook wasn’t listening through your phone microphone. Meta said it wasn’t tracking you across the internet. Meta said Cambridge Analytica wasn’t a problem. The company has a pattern of defining surveillance narrowly and deploying it broadly.

Employees are right to be skeptical. The rest of us should be paying attention. Because if Meta’s employees, people inside the surveillance machine, can’t opt out, what chance do the rest of us have?

References

Published: April 23, 2026