"Refining employees"? Meta collects employees' mouse movements and keyboard operations to train AI

Meta is currently turning its employees' daily computer operations into AI training data, pushing the tech giant’s bet on artificial intelligence to a new dimension—not only aiming to use AI to replace human labor, but first to "extract" human input to feed AI. On April 22, according to Reuters, an internal memo revealed that Meta has begun deploying tracking software on work computers of employees in the US, **collecting real-time mouse movements, clicks, and keyboard activity, as well as regularly taking screenshots, with all data used to train its AI models.** The tool is called the "Model Capability Initiative" (MCI), covering work-related applications and websites. **This deployment is part of Meta's broader "AI transformation" strategy.** Meta CTO Andrew Bosworth stated in another memo released Monday: **the company is ramping up internal data collection efforts, aiming to build AI agents that can "primarily undertake work tasks," while the role of human employees will shift to "commanding, reviewing, and assisting improvements."** The strategic project has been renamed the "Agent Transformation Accelerator" (ATA). ## Feeding AI with employees’ day-to-day operations: How the MCI tool works The memo shows that the MCI tool was announced internally on Tuesday by an AI research scientist from Meta's model building team, Meta SuperIntelligence Labs. The memo explained that the tool's goal is to fill gaps in how AI models simulate human-computer interactions—such as selecting options from drop-down menus or using keyboard shortcuts, actions current AI models still find difficult to naturally replicate. **"This is exactly where all Meta employees can help the model progress by their daily work," the memo stated.** Meta spokesperson Andy Stone confirmed that data gathered by MCI will be used as one input for AI training, and said this data will not be used for employee performance evaluation or any other purposes. Safeguards are in place to filter out "sensitive content," though he did not specify which types of data will be excluded. Stone further explained: > "If we are to build agents that help people accomplish everyday computer tasks, the models need genuine human operation samples—like mouse movements, clicking buttons, and navigating drop-down menus." Reportedly, **the deployment of MCI is a microcosm of Meta’s large-scale AI transformation strategy.** Internally, Meta has begun requiring employees to use AI agents for programming tasks, even if this results in reduced efficiency in the short term. Meanwhile, Meta is breaking functional boundaries between job roles, implementing a universal position called "AI builder." Last month, Meta set up a new Applied AI (AAI) engineering team, focused on improving AI models’ programming capabilities and using those models to build AI agents that can handle product development, testing, and release. Earlier this month, Meta began transferring "top-performing" software engineers to the AAI team. Andrew Bosworth’s vision in the memo: **"The future we are building is one where agents primarily take on the work, and our role is to command, review, and help them improve."** He added that the goal is for agents to "automatically recognize times when we need to intervene, so they do better next time." ## White-collar surveillance controversy: Legal boundaries and power imbalance This move has sparked obvious controversy regarding legal and labor rights. Yale law professor Ifeoma Ajunwa points out that computer logging and screenshot technology have historically been used by companies to investigate improper behavior or non-work activities by employees, but this time recording keyboard actions pushes data collection to a deeper level—subjecting white-collar workers to real-time monitoring previously reserved for delivery drivers and gig economy workers. **"At the federal level in the United States, there are no restrictions on employee monitoring," Ajunwa said. State laws at most require general notification when employers surveil employees.** In Europe, the legal landscape is entirely different. Valerio De Stefano, professor of technology and comparative labor law at York University in Toronto, said that such monitoring would likely be deemed illegal in Europe. In Italy, tracking worker productivity with electronic surveillance is explicitly prohibited; in Germany, courts have ruled that employers can only deploy keyboard logging in cases of suspected serious criminal offenses. Furthermore, De Stefano believes this approach may violate the EU’s General Data Protection Regulation (GDPR). De Stefano added that the expansion of employer surveillance awareness is changing the power dynamics in the workplace on a macro level, tilting the balance further in favor of employers.Risk Reminder and DisclaimerThe market involves risks, and investments should be approached with caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial circumstances, or needs of individual users. Users should consider whether any opinions, views, or conclusions discussed here fit their own situation. Any investment made based on this is at your own risk.