Your Clicks Are Now Meta’s Training Data
Meta has installed tracking software on U.S. employee computers — capturing every mouse movement, keystroke, and screen snapshot — to feed its AI agents. Here is exactly what the tool collects, why it exists, and what the law says about it.
Meta is tracking how its employees use computers — right down to which dropdown menus they click and which keyboard shortcuts they use. The data collected through a new internal tool called the Model Capability Initiative (MCI) will be fed directly into training AI agents that Meta hopes will one day handle significant chunks of office work on their own. According to internal memos reviewed and shared publicly, the software runs across a set of approved work apps and websites, logging clicks, mouse paths, and keystrokes — and takes periodic snapshots of what is on screen.
The announcement landed the same week Meta confirmed it plans to cut approximately 8,000 jobs — around 10 percent of its global headcount — starting May 20, 2026. The timing has not gone unnoticed inside the company. The MCI initiative sits inside a broader effort Meta is calling the Agent Transformation Accelerator (ATA), a rebranded version of its earlier “AI for Work” programme. The company is spending between $115 billion and $135 billion on AI capital expenditure in 2026 alone — nearly double what it spent in 2025.
“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.”— Andy Stone, Meta Spokesperson
Meta CTO Andrew Bosworth laid out the broader aim in a separate internal memo sent on Monday, April 20. The goal, as Bosworth described it, is a future where AI agents do the primary work and employees are there to direct and review them. He said Meta would be “rigorous” about collecting data and running evaluations across every type of work interaction. Meta spokesperson Andy Stone confirmed to reporters that MCI data would be one of those inputs, while also stating the data would not feed into performance reviews.
One anonymous employee told the BBC the situation felt “very dystopian,” given that the tool is arriving alongside expectations of deeper job cuts. Another recently departed staffer described MCI as “just the latest way they’re shoving AI down everyone’s throat.” Meta has not commented on either statement directly.
The MCI tool is not optional for affected employees. It runs passively in the background across a defined list of work apps and websites. According to the memo posted to Meta SuperIntelligence Labs’ internal channel, the software was described as a way for all employees to help “our models get better simply by doing their daily work.” The team credited with developing and deploying MCI works within Meta SuperIntelligence Labs, the AI division Meta restructured in 2025 after bringing in Alexandr Wang from Scale AI.
The workforce cuts tied to this timeline are substantial. Meta is planning to lay off approximately 8,000 employees on May 20, 2026 — representing 10 percent of its roughly 78,865-person global workforce. The company has indicated further cuts are likely in the second half of 2026, though timing and scale have not been finalised. California WARN Act filings confirm 124 positions at Meta’s Burlingame office effective May 22, and 74 at its Sunnyvale facility effective May 29.
Meta is also advertising far fewer open roles. A website the company uses to post jobs carried around 800 listings in March 2026. By April, that number had dropped to just seven. The company’s spokesperson declined to comment on the reduction when approached by the BBC.
The trend extends beyond Meta. Amazon has trimmed approximately 30,000 corporate employees in recent months — close to 10 percent of its white-collar workforce. Fintech company Block cut nearly half its staff in February 2026. The shift reflects a broader pattern among large U.S. tech companies using AI tooling as justification for workforce reductions. Meta, for its part, has been asking employees to use AI coding agents even when doing so slows them down in the short term, according to internal communications described to Reuters. You can read our coverage of how tech leadership transitions are playing out across the industry, and how tools like those from hardware brands are navigating the same AI-driven market.
“On the U.S. side, federally, there is no limit on worker surveillance. State-level laws require at most that workers be broadly informed when employers are monitoring them.”— Ifeoma Ajunwa, Law Professor, Yale University
Computer logging and screen-capture software has historically been used by employers to investigate misconduct or monitor productivity in hourly roles. Keystroke logging of this kind — applied at scale across a white-collar technical workforce — is different in scope. York University law professor Valerio De Stefano, who researches technology and labour law comparatively, said awareness of employer surveillance shifts workplace power dynamics in the employer’s favour. His remarks to Reuters referred specifically to European legal frameworks, where the practice would face significant legal barriers. You can track related debates on how tech platforms are changing user and worker experiences across the board.
Meta’s MCI tool, the associated Agent Transformation Accelerator programme, and the May 20 workforce reduction were covered above. The tracking software applies to U.S.-based employees. The data it collects — mouse movements, clicks, keystrokes, and periodic screen snapshots — is being used to train AI agents within Meta SuperIntelligence Labs, the division led by Alexandr Wang following Meta’s $14.3 billion Scale AI investment in 2025. Meta’s AI capex for 2026 stands at $115–135 billion. Legal experts cited in the reporting noted that equivalent surveillance practices would face significant restrictions under European law, including GDPR, and are explicitly prohibited in countries such as Italy. The first model from Meta Superintelligence Labs, Muse Spark, was released in early April 2026. Further reading on adjacent topics: upcoming hardware launches and technology in broader contexts are covered in our recent stories.
