Meta Platforms Inc. has begun tracking the granular digital movements of its employees, including every keystroke and mouse click — to feed the hungry maw of its artificial intelligence (AI) training programs.
The social media giant, which owns Facebook and Instagram, informed staff on Tuesday that a new internal tool, dubbed the Model Capability Initiative (MCI), will now log activity across company computers and applications, sparking internal backlash over privacy concerns.
The software is designed to capture screen snapshots and monitor how workers navigate dropdown menus and use keyboard shortcuts to provide “real-world examples” for autonomous AI agents, according to internal memos.
A Meta spokesperson defended the initiative, insisting the company intends to build AI agents capable of completing everyday tasks, but the models must first observe how humans perform them.
“The data is not used for any other purpose,” the spokesperson said, emphasizing safeguards are in place to protect sensitive content and that the data will not influence performance reviews.
While Meta insists the MCI tool is strictly for model training, specifically for its recently launched Muse Spark model, critics argue that the line between data collection and employee surveillance is becoming increasingly blurred. The strategy may be old, but they fear AI tools will allow for faster, more ruthless cost cutting and job displacement on a far wider scale now.
“This type of idea represents an enormous privacy and ethics engineering infrastructure and an active continuous governance system to even hope to hit the high notes in places like California with both constitutional and statutory specific employee protections,” said Michelle Finneran Dennedy, chief experience officer at Lokker Inc.
What is more, the timing of the rollout has left many employees feeling uneasy. The initiative arrives amid a climate of extreme job insecurity within the company. Meta has already eliminated approximately 2,000 positions this year, and reports suggest a much larger wave of layoffs — potentially affecting 10% of the global workforce, or 8,000 people — is slated for late May.
“This company has become obsessed with AI,” one employee told the BBC, describing the micro-tracking as “very dystopian” while the specter of further job cuts looms.
The deployment of MCI is a direct result of CEO Mark Zuckerberg’s aggressive adoption of Superintelligence. Meta is projected to spend roughly $140 billion on AI in 2026, nearly doubling its investment from the previous year. This capital flight into silicon and software is visible elsewhere; the company’s job board, which hosted 800 listings in March, has reportedly dwindled to just seven active openings.
Zuckerberg previously signaled that 2026 would be the year AI “dramatically changes the way we work,” noting that tasks once requiring large teams are now being handled by a single person equipped with the right tools.
“The conversation around Meta tracking employee movements tends to focus on the surveillance dimension. That’s understandable, but it misses the most technically significant point,” Prevalent AI CEO Paul Stokes said. “What makes this data interesting isn’t that it tells you what employees did. It’s that it begins to capture how and why decisions were made.”
Meta’s move also reflects a broader, more aggressive trend in corporate “work surveillance.” Similar technologies have been deployed at firms like JPMorgan, where keystrokes and video call durations are monitored to track productivity.
“Big Brother is Watching You has become Big LLM is Watching You,” said A.K. Pradeep, founder and CEO of Sensori.ai. “Taylor and the Gilbreths gave us time-and-motion studies that transformed industrial productivity. LLM observation of human interaction could do the same and will have many beneficial consequences – structuring better UI, understanding the overt and covert workflows, the bits and pieces of work that make people pause, struggle, create, and edit.”
The shift toward behavioral harvesting converts individual human expertise into proprietary code, creating a paradox where workers could actively train the very agents that can augment their work experience as well as potentially replace them, added Ron Westfall, an analyst at HyperFRAME Research.
“The move raises key privacy alarms, as intensive monitoring inevitably blurs professional lines and risks the accidental collection of sensitive personal or proprietary data,” Westfall said in a message. “While this approach may accelerate the development of autonomous AI, I see it risking a dystopian breakdown in company culture by prioritizing technical benchmarks over the trust and psychological well-being of the workforce.”
For a workforce already bracing for a 10% reduction in headcount, being asked to “train their replacements” by simply doing their daily work has created a palpable sense of friction at Meta.
Said Dr. Seth Dobrin, CEO and founder of Arya Labs: “This is the logical endpoint of a data philosophy that was never going to stop at the public internet. Meta scraped the web without consent. Now they are scraping their own workforce. The pattern is consistent: treat data as found material, not as something people produce with rights attached. None of it is necessary. If your thesis is that more data produces a better model, you will always need more data, and you will always find it in places where the people generating it cannot push back. Public users could not. Employees cannot either.”
