Meta is recording what its US employees do on their computers and using that data to train its AI agents.
The company has deployed a tool called Model Capability Initiative (MCI) on US-based employee machines. According to Reuters, MCI runs during work sessions capturing mouse movements, clicks, keystrokes, and periodic screenshots. That data feeds directly into Meta's AI agent training pipeline. The Verge reported on the program based on Reuters' original coverage. Meta has not published details about data retention periods or whether employees had any choice about participating.
What Meta Is Actually Trying to Capture
The move reflects a real problem in AI development: public web data for training large language models (the AI systems behind chatbots) is increasingly scraped out or tied up in legal disputes. Behavioral data from actual workers is a different category. It captures how people navigate software, the exact sequence of steps they take to finish real tasks, and the dead ends they hit along the way. That implicit, procedural knowledge is hard to source from text on the internet - and it is precisely what you need to train AI agents, which are systems that don't just answer questions but carry out multi-step tasks on your behalf.
The Consent Problem
Consumer apps ask you to click "agree." Employees generally cannot decline software installed by their employer. That gap matters.
Microsoft and Google both collect usage telemetry to improve AI features, but passive recording via screenshots and keystrokes goes considerably further than standard product analytics. Several European countries require employee representatives to formally approve monitoring software before deployment. US law offers workers much thinner protection.
Meta's approach may genuinely advance its AI agent capabilities - real work patterns are a richer training signal than curated datasets. It also sets a precedent other large employers could follow: treating their own workforce as a live training dataset, regardless of whether employees knew that was part of the job.