Meta is training AI on employee data while laying off thousands

Meta has been recording what its employees type, click, and do on their work laptops—and feeding that data directly into AI training. The program, called the Model Capability Initiative (MCI), was rolled out in April 2026 and captures keystrokes, mouse movements, and screenshots from US-based staff. The same week 8,000 employees received layoff notices, leaked audio emerged of Mark Zuckerberg defending the surveillance program and confirming there is no opt-out.

The program

MCI tracks activity across tools employees use every day: Google, LinkedIn, GitHub, Slack, and Wikipedia, per CNBC. Zuckerberg's argument, heard on the leaked audio, is that the average Meta employee is far more skilled than any outside contractor the company could hire for data-labeling work—so capturing how they actually work produces better training data. CTO Andrew Bosworth confirmed on record that opting out on corporate laptops is not an option.

Most tech giants use low-cost, outsourced annotation workers to label AI training data. Meta's pitch is that quality beats quantity. The irony is stark: the better you are at your job, the more useful your behavior is for training a model that might eventually replace the role.

The pushback

More than 1,000 employees signed an internal petition against MCI, according to TechTimes. UK-based staff went further, launching a formal union drive with United Tech and Allied Workers—the first organized labour response to AI-driven workplace surveillance at a major US tech company.

The layoffs arriving in the same week as the audio leak sharpened the tension. Meta began notifying roughly 8,000 employees—about 10% of its global workforce—of terminations on May 20, 2026. Median employee compensation had already dropped from $417,400 in 2024 to $388,200 in 2025, with further stock-raise cuts in February 2026. The company is spending $125–145 billion on AI this year.

The EU carve-out

European employees are exempt from MCI. GDPR bars keystroke surveillance without explicit consent, so Meta simply didn't roll the program out there. That means the AI being built on this data is trained entirely on US work patterns—a legal shortcut with a geographic boundary drawn around it.

No US data protection law creates an equivalent barrier for American workers. Until Congress acts or more employees organize, MCI stays in place.