Meta to capture employee interaction data
- Reports say Meta will capture employees' mouse movements, clicks, keystrokes, and occasional screen snapshots on work devices. - The company plans to use those interaction traces as training data for internal AI systems. - The surveillance plan has sparked privacy and labour concerns about workplace data collection for model training. (rappler.com)
Meta is installing software on U.S. employees’ work computers to record clicks, keystrokes, mouse movements and screen snapshots for AI training. (reuters.com) The tool is called Model Capability Initiative, or MCI, and Reuters reported on April 21 that Meta told staff it would run across work apps and websites as the company builds AI agents that can complete office tasks on their own. (reuters.com) CNBC reported on April 22 that the monitoring list includes Google, LinkedIn and Wikipedia, along with hundreds of other sites and apps used on company devices. (cnbc.com) The underlying idea is simple: an AI agent learns from examples of how people move through software, the way a trainee learns by watching a coworker click through a task step by step. Meta is collecting those traces from employees instead of relying only on public web text or synthetic data. (techcrunch.com) That reflects a wider problem in AI development: text models can answer questions, but software agents need records of real human actions inside browsers, forms and workplace tools. Reuters said Meta described the program as part of a broader push to build systems that can perform work tasks autonomously. (reuters.com) The labor and privacy concern is not just that Meta can watch work devices, but that the same activity is being repurposed as raw material for model training. CNBC reported that internal messages showed employee concern about how to opt out and what exactly the system would capture. (cnbc.com) Meta has not invented workplace monitoring, but this plan ties surveillance directly to product development. TechCrunch noted that the company is turning routine mouse movements and button clicks into data for internal AI systems, extending a practice common in productivity software into model training. (techcrunch.com) The legal backdrop is also tightening outside the United States. The European Union’s AI Act, which entered into force in 2024, bans some workplace emotion-recognition uses and classifies other employment-related AI systems as high risk, putting more scrutiny on how employers deploy AI around workers. (eur-lex.europa.eu) Meta’s immediate next test is not technical but organizational: whether employees accept a system that treats everyday computer use as training data. For now, the company’s AI push is moving through the keyboard and mouse of its own workforce. (reuters.com)
