Meta Is Recording Every Keystroke Its Employees Make

The Surveillance Company Turned the Cameras Inward

On April 21, 2026, Reuters revealed that Meta had quietly begun deploying monitoring software on its US employees' work laptops. The program, internally called the Model Capability Initiative, captures keystrokes, mouse movements, clicks, and takes periodic screenshots of whatever employees are doing on their screens ^[1].

Every dropdown menu selection. Every keyboard shortcut. Every message typed in Slack. Screenshots of whatever's on your screen when the system decides to snap one.

Meta says the data feeds directly into its AI training pipeline. The company wants to build AI agents that can do office work the way humans do, and the fastest way to teach a machine how humans click through software is to record humans clicking through software ^[1][2].

A Meta spokesperson told Reuters: "If we're building agents to help people complete everyday tasks using computers, our models need real examples of how people actually use them" ^[1].

That framing makes it sound reasonable. The reality is less polished.

What's Being Tracked

CNBC reported on April 22 that the list of monitored websites and apps runs into the hundreds ^[3]. The confirmed targets include:

• Google (search, Gmail, Docs, including personal email if accessed on work devices)
• LinkedIn
• Wikipedia
• GitHub (Microsoft-owned)
• Slack (Salesforce-owned)
• Atlassian (Jira, Confluence)
• Meta's own properties (Threads, Workplace, Manus)

Think about what flows through those tools on a normal workday. Code commits with proprietary logic. Slack messages about medical leave. Google searches about job openings at other companies. LinkedIn messages to recruiters. Personal emails on Gmail.

All of it now feeds into Meta's AI training data.

No Opt-Out. Period.

When employees asked CTO Andrew Bosworth whether they could decline, he was direct: "No there is no opt out on your work provided laptop" ^[4].

The only escape is geographic. European employees are exempt because the EU's General Data Protection Regulation (GDPR) and national worker protection laws make this kind of monitoring illegal without explicit consent ^[4][5]. US employees get no such protection.

That's the gap in a sentence. European privacy law says you can't record your employees' keystrokes without consent. American privacy law says nothing at all.

Employees Called It "Dystopian"

Multiple Meta employees described MCI as "dystopian" to reporters, according to CNBC ^[3]. Their concerns weren't abstract:

• Passwords: Keystroke logging captures everything typed, including credentials for non-Meta services
• Sensitive personal data: Immigration status, health information, and family details discussed in work tools
• Product secrets: Screenshots could capture unreleased product details, code, and internal strategy documents
• Data misuse: No clear guardrails on who accesses the training data or how long it's retained

Meta claims safeguards exist to protect sensitive content, but employees reported that the company hasn't been transparent about what those protections actually are ^[4].

Training Your Replacement

Platformer's Casey Newton framed MCI as "Taylorism" applied to knowledge work: the century-old practice of breaking human labor into measurable, reproducible steps so it can be systematized and eventually automated ^[4].

The timing makes the point better than any metaphor could. On April 23, two days after Reuters broke the MCI story, Meta announced it was cutting 10% of its workforce (roughly 8,000 employees) plus eliminating 6,000 unfilled positions ^[4][6].

Mark Zuckerberg told employees that "single talented individuals" could now accomplish work that previously required large teams. The mechanism for that accomplishment? AI agents. Trained on the keystrokes of the people being replaced ^[6].

Bosworth's vision is explicit. He wrote that a future is coming where AI agents "primarily do the work" while employees "direct, review and help them improve" ^[4]. The employees recording their own keystrokes right now are building that future, one in which fewer of them are needed.

The $14 Billion Data Pipeline

MCI didn't appear in a vacuum. In February 2025, Meta acquired a 49% stake in Scale AI for $14.3 billion and hired its CEO, Alexandr Wang, to lead a new division called Meta Superintelligence Labs ^[1][4].

Scale AI's core business is preparing training data for AI models. Wang's mandate at Meta is to close the gap with OpenAI, Anthropic, and Google in generative AI capabilities. The problem: those competitors already have access to massive amounts of training data through their consumer products. Meta needed its own source ^[1].

Its employees became that source.

Every keystroke, every mouse path through a complex Jira workflow, every copy-paste between tools: it all becomes training data for AI agents that Meta plans to sell to enterprises. The workers generating the data don't get a cut. They get a monitoring notice and a reminder that their work laptop isn't really theirs.

Why This Matters Beyond Meta

If you don't work at Meta, here's why this should bother you.

Meta employs roughly 72,000 people. That's 72,000 people whose work habits, communication patterns, and digital behavior are being harvested to train AI agents. Those agents won't stay inside Meta. They'll be packaged and sold to every enterprise that wants to automate knowledge work ^[4].

And if Meta gets away with it (no legal challenge, no regulatory action, no employee revolt that sticks) every other tech company will follow. Amazon, Google, Microsoft, and every startup with AI ambitions will have a template for turning their employees into involuntary training data generators.

We've spent a decade tracking how tech companies surveil their users. Now they're surveilling their own workers to build the machines that will surveil everyone else.

What You Can Do

If you work at a company deploying similar monitoring tools:

• Never access personal accounts on work devices. No personal email, no banking, no medical portals. Assume everything on the work laptop is being recorded.
• Use a personal phone for private communications. Keep work devices for work only.
• Check your jurisdiction. If you're in the EU, California, Illinois, or other states with biometric or workplace privacy laws, your employer may need your consent before deploying keystroke loggers.
• Ask your IT department directly: "Does our company use any keystroke logging, screen recording, or mouse tracking software on employee devices?" Get the answer in writing.
• Talk to coworkers. Collective awareness is the first step toward collective action. Meta employees pushed back internally, even if Bosworth shut them down.

The Bottom Line

Meta built the world's most sophisticated people-tracking infrastructure and pointed it at 3 billion users. Now it's pointing that same instinct inward, at its own employees, to build AI that can do their jobs.

The company that knows what you like, who you message, and where you go now also knows exactly how its employees move a mouse, type a password, and navigate a spreadsheet. That knowledge isn't staying inside Meta. It's becoming the foundation for AI agents that every corporation will use.

Your coworker's replacement might be trained on someone else's keystrokes. In a few years, that might be the only kind of training data there is.