Meta‘s employee monitoring program, called the Model Capability Initiative, has begun logging every mouse movement, click, and keystroke from U.S. staff computers. The goal is to train AI models to carry out the same tasks. The program also takes occasional screenshots, capturing how people use software so AI systems can learn to replicate that behavior.
The initiative is part of Meta’s Agent Transformation Accelerator, a push to build AI capable of handling routine tasks across different tools and platforms. An internal memo described the rationale: “This is where all Meta employees can help our models get better simply by doing their daily work.” The program was first reported by Reuters.
What Meta employee monitoring captures
Text data alone cannot teach an AI model how to navigate a menu, move between windows, or parse different website formats. These are the small, practical interactions that define a workday but rarely get documented in any structured way. Meta’s employee monitoring program captures them directly, turning routine activity into training data for AI agents.
The level of detail collected is more common in factories and warehouses than in offices. Logging every mouse movement and keystroke gives Meta’s AI models a granular picture of how employees actually complete tasks. Meta says the data is not used for performance evaluation, and that safeguards protect sensitive information.
The timing alongside workforce cuts
Meta’s employee monitoring rollout coincides with plans to cut roughly 10% of the company’s global workforce, with further reductions expected later in 2026. The company is capturing behavioral data from the same employees it’s laying off, to train the AI that may eventually perform those roles.
The program is U.S.-only so far. European labor and data privacy regulations impose strict limits on this kind of workplace surveillance. In the U.S., companies are generally required only to inform employees that monitoring is in place.
Why other companies will likely follow
Meta’s employee monitoring approach addresses a practical gap in AI agent development. As AI systems become more capable of performing tasks rather than just generating text, behavioral training data becomes more useful than scripted demonstrations. Meta has a large workforce and the legal environment to collect it at scale.
If the program delivers what Meta expects, other organizations will face the same reasoning. Any company building AI agents for office work could benefit from capturing how employees actually complete tasks. Whether they choose to act on that is another question.
Meta’s stated goal is AI that can fully replicate what human employees do at a computer. Whether those models become tools that assist workers or systems that replace them depends entirely on how they’re deployed.