Meta logs employee activity
- Reports say Meta plans to record employee mouse movements, clicks, and keystrokes to train internal AI agents. - Multiple Spanish outlets report Meta aims to capture 'every action' of office workers to build training data for agent systems. - Coverage highlights privacy and labor backlash risks as companies use internal activity to bootstrap agent training. ( )
Meta is installing software on U.S. employees’ work computers to record mouse movements, clicks and keystrokes for training its artificial intelligence agents. (reuters.com) Internal memos reviewed by Reuters said the tool is called Model Capability Initiative, or MCI, and that it will run across work-related apps and websites. Reuters reported the system can also take occasional screenshots of what is on workers’ screens. (reuters.com) Meta confirmed the project to TechCrunch and said the goal is to give its models “real examples” of how people use computers, including mouse movements, button clicks and dropdown navigation. The company said the data is collected on “certain applications” and is “not used for any other purpose.” (techcrunch.com) The basic idea is simple: an AI agent is software meant to carry out computer tasks the way a person would, such as clicking through menus, filling forms or moving between apps. Training one requires examples of those step-by-step actions, not just text scraped from the web. (techcrunch.com) CNBC reported the tracking list includes hundreds of websites and apps, among them Google, LinkedIn, Wikipedia, GitHub, Slack and Atlassian products. Its report said an earlier internal list also included OpenAI’s ChatGPT and Anthropic’s Claude before the list changed. (cnbc.com) The project sits inside a broader push by Chief Executive Mark Zuckerberg to speed up Meta’s artificial intelligence work after the company spent heavily on new talent and new models. CNBC said the effort is tied to Meta Superintelligence Labs, and Meta said this month that its new Muse Spark model is the first in a new series from that unit. (cnbc.com, about.fb.com) Meta has also been expanding AI inside its own operations, not just in consumer products. In January, the company said 2026 would be a year when AI would “transform how we work,” and in March it said AI had been integrated into its internal risk review process for privacy, safety and security checks. (about.fb.com, about.fb.com) The immediate dispute is not about whether Meta is building agents, but about what data it is using to do it. Reuters and CNBC both reported that the company circulated internal messages after workers raised concerns about surveillance and privacy on internal chat boards. (reuters.com, cnbc.com) Meta’s public line is that the tracking is for model training, with safeguards around sensitive content, rather than for employee performance reviews. The test for the company now is whether workers accept that distinction once the software is running on their machines. (techcrunch.com, reuters.com)
