Meta employee‑data plan
- Reports say Meta built an internal tool to capture employee behaviour on work apps and websites for AI training. - Coverage warned the approach risks exposing sensitive information and raising regulatory concerns. - The story underscores that telemetry for workplace agents can trigger privacy backlash without clear safeguards ( )
Meta is installing software on U.S. employees’ work computers to capture clicks, mouse movements and keystrokes for AI training. (usnews.com) Reuters reported on April 21 that the tool, called Model Capability Initiative, was described in internal memos as part of a push to build AI agents that can carry out office tasks on their own. Meta told staff the software would run on company devices used by U.S.-based employees. (usnews.com) CNBC reported on April 22 that Meta planned to monitor activity across hundreds of apps and sites, including Google, LinkedIn and Wikipedia, using internal documents it reviewed. Those records said the system could log keystrokes and mouse clicks as workers moved through common software. (cnbc.com) The basic idea is imitation learning: collect examples of how people finish digital tasks, then use those traces to teach software agents to repeat them. In this case, the “examples” are employees navigating menus, typing into fields and switching between work tools on live systems. (computerworld.com) Meta said in the Reuters report that the data would be used to train AI models, not to evaluate worker performance. Internal memos also said the company would exclude some sensitive categories, though the reports said employees still raised concerns about what could be captured on screen. (usnews.com) Those concerns center on screen context as much as keystrokes. Reuters said the software could collect screen content, and coverage in CNET and Entrepreneur said periodic screenshots or on-screen snapshots were part of the system described to employees. (cnet.com) The plan lands as Meta is spending heavily to expand its AI business. In its 2026 strategy post, the company said AI would “transform how we work,” and Meta has guided to $115 billion to $135 billion in 2026 capital expenditures, according to earnings coverage and company materials. (about.fb.com) Meta has already been pushing AI deeper into its products and internal operations. The company said this month that its Muse Spark model powers the Meta AI app and will roll out across WhatsApp, Instagram, Facebook, Messenger and AI glasses in the coming weeks. (about.fb.com) The employee-tracking project also arrives after earlier fights over AI training data. TechCrunch reported in 2025 on lawsuits over Meta’s use of copyrighted books, and Meta’s own posts have described using public content and user interactions in some products to improve AI systems. (techcrunch.com) What happens next will depend on how narrowly Meta defines “work apps,” how long it keeps the data and what employees can opt out of on company machines. For now, the company’s own staff have become one more source of training data in the race to build AI that can do desk work. (cnbc.com)
