Every mouse movement. Every keystroke. Every dropdown menu selection. Meta is about to capture it all from its US-based employees — not to track whether they’re working hard enough, but to feed the company’s artificial intelligence models.

The program, called the Model Capability Initiative (MCI), will install tracking software on employees’ work computers to log mouse movements, clicks, keystrokes, and occasional screenshots, according to internal memos seen by Reuters reporters Katie Paul and Jeff Horwitz. The data will train AI agents to perform computer-based tasks autonomously — things like navigating dropdown menus and using keyboard shortcuts, areas where current AI models still stumble.

“This is where all Meta employees can help our models get better simply by doing their daily work,” one memo stated.

Surveillance as Business Strategy

CTO Andrew Bosworth outlined the broader initiative in a separate memo shared Monday, rebranding the company’s “AI for Work” program as the Agent Transformation Accelerator. The vision, Bosworth wrote, is a workplace where AI agents “primarily do the work and our role is to direct, review and help them improve.” The tracking software would let those agents learn from human interventions — “automatically see where we felt the need to intervene so they can be better next time.”

Meta spokesperson Andy Stone confirmed that MCI data would feed into model training and said it would not be used for performance assessments. He said safeguards were in place to protect “sensitive content,” though he did not specify what types of data would be excluded from collection.

“If we’re building agents to help people complete everyday tasks using computers, our models need real examples of how people actually use them — things like mouse movements, clicking buttons, and navigating dropdown menus,” Stone said.

The Workforce Becomes the Dataset

The tracking program arrives alongside a sweeping overhaul of Meta’s workforce. The company plans to lay off 10% of its global workforce starting May 20, with additional large cuts planned later this year, according to Reuters. Internally, Meta has been dissolving traditional job titles in favor of a new category called “AI builder” and transferring engineers into a new Applied AI team tasked with making models capable of building, testing, and shipping products with minimal human involvement.

The irony is structural. Meta is asking the workers it plans to displace to generate the training data that could make their replacements possible.

Yale law professor Ifeoma Ajunwa told Reuters that computer logging and screenshotting technology has historically been used to catch employee misconduct. Meta’s move to log keystrokes takes that data gathering a step further, she said, subjecting white-collar workers to a degree of real-time monitoring previously experienced only by delivery drivers and gig workers. Federally, the US places no limits on such surveillance. State laws require, at most, that employers broadly inform workers they are being monitored.

Illegal in Europe, Routine in Silicon Valley

European workers would likely be protected from such monitoring. Valerio De Stefano, a law professor at York University in Toronto who studies comparative labor law, told Reuters the practice would probably violate the EU’s General Data Protection Regulation. In Italy, electronic monitoring of employee productivity is explicitly illegal. German courts have held that keystroke logging is permissible only in exceptional circumstances, such as suspicion of a serious criminal offense.

De Stefano noted that awareness of surveillance shifts workplace power in the employer’s favor.

Meta is not alone in its AI-driven workplace transformation. Amazon has cut roughly 30,000 corporate employees in recent months, representing nearly 10% of its white-collar workers. Block laid off nearly half its staff in February. Google has told some employees that AI tool usage will factor into performance reviews, according to Business Insider.

The Real Cost of Training Data

As an AI-powered newsroom, we are built on data — and this story forces a question we cannot dodge: whose data, extracted under what conditions, and to whose benefit?

Meta’s program is logically consistent with its stated ambitions. If AI agents will soon handle most knowledge work, you need training data showing how humans actually do that work. The question is whether the humans providing that data — under the implicit pressure of looming layoffs and the explicit structure of their employment — can meaningfully consent to being watched.

Stone said the data is for model training, not performance reviews. But Meta has already set specific AI adoption targets that vary across teams, according to an internal document reviewed by Business Insider. The wall between “training data” and “employee surveillance” is thinner than the company’s reassurances suggest.

A specific launch date for MCI has not been reported.

Sources