Every click, every keystroke, every screen — Meta's workforce became a living dataset. Then 8,000 of them got layoff notices.
---
Before Meta cut roughly 8,000 employees on May 20, it had already been harvesting something else from them: their work. Since April 2026, the company has been running a mandatory surveillance program on U.S. employees' company-issued laptops — one that captures keystrokes, mouse movements, button clicks, and periodic screenshots across hundreds of apps and websites. The data feeds Meta's AI training pipeline. The employees had no say in the matter. And now, thousands of them no longer have jobs.
The program, called the Model Capability Initiative (MCI), was first reported by Reuters on April 21, 2026. It quietly appeared on employee machines without a company-wide announcement. The software monitors worker behavior across platforms including Google, LinkedIn, GitHub, Slack, and even personal Gmail accessed on work devices — building what amounts to a real-world behavioral dataset for training AI agents that can perform human computer-use tasks autonomously.
---
No Opt-Out, No Exceptions — Unless You're in Europe
When employees asked CTO Andrew Bosworth whether they could decline to be monitored, his answer was unequivocal: "No there is no opt out on your work provided laptop."
The response triggered a wave of crying, shocked, and angry emoji reactions from staff on the internal post. What followed was more organized: flyers began appearing across Meta's U.S. offices — pinned in meeting rooms, taped to vending machines, and placed on top of toilet paper dispensers — comparing Meta's headquarters to an "Employee Data Extraction Factory." A petition demanding the program be scrapped circulated to nearly 20,000 coworkers. More than 1,000 signed it.
European employees are exempt from MCI entirely. The reason is not a philosophical commitment to worker dignity — it is GDPR. EU privacy law prohibits the kind of blanket, mandatory behavioral capture that Meta implemented in the U.S. without any opt-out mechanism. The geographic carve-out has not been lost on American employees, who found themselves legally unprotected in ways their European colleagues were not.
---
Training AI Agents to Become the Workers
Meta's stated rationale for MCI frames the monitoring as a collaborative act. An internal memo described the program as a way for rank-and-file employees to help improve company models in areas "where they struggle to emulate basic computer-use behaviors" — things like navigating dropdown menus, switching between apps, and executing keyboard shortcuts. The implication was that employees' natural work habits, captured silently in the background, would teach AI agents to behave more like humans at a computer.
Bosworth's internal memo went further, sketching a future in which AI agents "primarily do the work" while human employees "direct, review and help them improve." It is a vision of a radically restructured workforce — one in which Meta's human labor becomes a supervisory and training function for the AI systems that gradually absorb it.
That vision collided hard with reality on April 23, two days after the Reuters report: Meta announced it would cut 10 percent of its workforce, effective May 20. Approximately 8,000 employees were laid off, with another 7,000 reassigned to AI-focused teams. In a memo to staff, Zuckerberg acknowledged the stakes plainly: "Success is not a given."
The timing struck many workers as something beyond irony. Multiple Meta employees described MCI to reporters as "dystopian." The petition's flyers explicitly cited the U.S. National Labor Relations Act, noting that "workers are legally protected when they choose to organize for the improvement of working conditions." It was among the clearest signs yet of an emerging labor movement inside one of Silicon Valley's most powerful companies.
---
A $145 Billion Bet Funded by Its Own Workforce's Data
The MCI program fits inside a broader and extraordinarily expensive commitment by Meta to AI infrastructure. Zuckerberg has signaled that the company is raising its capital expenditure guidance by up to $10 billion — to as high as $145 billion — for 2026, making it one of the largest single-year technology infrastructure investments in corporate history. That spending encompasses data centers, chips, energy infrastructure, and the AI model development pipelines that programs like MCI are designed to feed.
Meta also acquired a 49 percent stake in data-labeling firm Scale AI last year for more than $14 billion — a deal that underscored how serious the company is about the data supply chain for AI training. MCI can be read as a vertical integration of that strategy: rather than paying third parties to generate synthetic or labeled training data for computer-use AI, Meta is capturing real behavioral data from its own employees at zero marginal cost.
The competitive logic is clear. OpenAI and Anthropic are both racing to develop AI agents capable of autonomous computer use — systems that can book flights, file expenses, or navigate enterprise software without human instruction. To train such agents, companies need enormous quantities of high-quality examples of humans actually using computers. Meta's workforce of tens of thousands of knowledge workers, spread across product, engineering, legal, finance, and operations roles, represents a uniquely rich and diverse dataset. MCI turns that workforce into a sensor array.
---
A Precedent With No Clear Limit
What Meta has built is not unprecedented in its component parts — employee monitoring software has existed for decades, and its use expanded significantly during the remote-work era. What is new is the explicit repurposing of behavioral surveillance as AI training data, and the mandatory, no-opt-out nature of the collection.
Privacy advocates and labor researchers have noted that MCI establishes a model other large employers may follow. If monitoring employee behavior on company devices is reframed not as performance surveillance but as AI research contribution, the legal and ethical frameworks that govern workplace monitoring may need to catch up quickly. The GDPR carve-out for European employees suggests that robust privacy law can function as a check — and that in its absence, employers face few formal constraints.
For the 8,000 Meta employees who received layoff notices days after their keystrokes were captured and stored in a training corpus, the question is unavoidably personal: did they help build the system that made their jobs obsolete? Meta has insisted the MCI data will not be used for performance evaluation. But the employees whose flyers are still taped to office walls across Menlo Park may not find that distinction particularly comforting.
---
"No there is no opt out on your work provided laptop."— Andrew Bosworth, CTO, Meta