Workplace Keystroke Tracking

Meta is installing software on U.S.-based employees’ work computers to capture keystrokes, clicks, mouse movements and occasional screen snapshots for artificial-intelligence training. (reuters.com) The software is called Model Capability Initiative, or MCI, and Reuters reported on April 21 that Meta described it in internal memos seen by the news agency. Meta spokesperson Andy Stone said the data would be used for model training, not performance reviews. (reuters.com) Meta said the tool will run inside work-related apps and websites because its models still struggle with routine computer actions such as choosing from dropdown menus and using keyboard shortcuts. A Meta statement to TechCrunch said the company needs “real examples” of mouse movements, button clicks and navigation to build agents that can help people complete everyday computer tasks. (reuters.com) (techcrunch.com) That is the basic idea behind this kind of system: an artificial-intelligence “agent” is software meant to operate a computer the way a person does, by clicking, typing and moving through menus. To train one, a company needs examples of those actions, much like a driving system needs footage of real roads. (techcrunch.com) (reuters.com) The rollout sits inside a broader internal push to remake work around AI. Reuters reported that Chief Technology Officer Andrew Bosworth told employees in a separate memo that Meta’s “AI for Work” effort had been rebranded as Agent Transformation Accelerator, with more internal data collection to follow. (reuters.com) Bosworth wrote that Meta’s goal is a workplace where agents “primarily do the work” and employees direct, review and improve them. Reuters reported that Meta linked the new tracking to that plan, saying agents should learn from the moments when workers intervene on a task. (reuters.com) The privacy tension is unusually direct because the training data comes from employees, not from public websites or customer prompts. Reuters reported that Meta said safeguards would protect “sensitive content,” but the company did not publicly spell out which categories of information would be excluded. (reuters.com) That cuts against Meta’s public message that privacy protections are built into its operations. On its own website, Meta says it has invested heavily in privacy since 2019 and that privacy teams are embedded across product groups, with twice-yearly privacy roadmapping built into product plans. (meta.com) Meta has been pressing employees to use more AI in daily work for months. 404 Media reported in October 2025 that a Meta vice president told metaverse teams to use AI to “go 5x faster,” and Business Insider reported in March 2026 that some groups were given explicit adoption goals for AI tools. (404media.co) (businessinsider.com) Meta has not said publicly that it will change course on MCI. For now, the company is treating ordinary office work itself as training data for the next generation of workplace AI. (reuters.com)