Meta tracks employee keystrokes and screenshots to train AI, sparking privacy backlash
Meta has rolled out a new internal tool that tracks employees' mouse movements, keystrokes, and even occasional screenshots on their work computers, using this data to train its artificial intelligence models. The software, dubbed the "Model Capability Initiative," is being deployed on all US-based employees' machines and runs on work-related apps and websites, according to internal memos reported by Reuters and shared by TechCrunch. A Meta spokesperson explained that the goal is to provide "real examples of how people actually use" computers—like navigating dropdown menus or using keyboard shortcuts—to build AI agents capable of handling everyday tasks autonomously.[web:1]
The initiative has sparked significant backlash among Meta staff, who view it as an invasive form of surveillance that raises serious privacy concerns. Employees are reportedly "up in arms," as described by Business Insider, over the mandatory program that captures intimate details of their daily computer interactions without opt-out options. Internal discussions, including posts in Meta's dedicated channels for its model-building team, highlight fears that the data could expose sensitive information such as performance metrics, personal identifiable information from HR systems, customer data, or even passwords typed repeatedly during the workday.[web:3]
Meta justifies the tracking as essential for advancing its AI ambitions, emphasizing that "all Meta employees can help our models get better simply by doing their daily work." The company insists safeguards are in place to protect sensitive content, though details remain sparse, with spokesperson Andy Stone declining to elaborate beyond general assurances. This move aligns with broader efforts to replicate human-computer interactions for more efficient AI, but critics like Yale University law professor Ifeoma Ajunwa note that US federal law imposes no limits on worker surveillance, unlike stricter European regulations that would likely prohibit such practices.[web:2]
The timing adds to employee unease, as Meta plans to slash 10% of its global workforce starting next month, with more cuts to follow later this year. This combination of heightened monitoring and layoffs fuels perceptions of a "dystopian" shift toward automating jobs, potentially affecting thousands of workers who now contribute unwittingly to the very technology that could replace them. For those involved, the immediate impact includes adjusted workflows on monitored apps, while the data fuels Meta's push into AI agents designed for autonomous task completion.[web:2]
Looking ahead, questions linger about data security, with online discussions warning that such granular logs—keystrokes, movements, and screenshots—pose risks of breaches that could leak sensitive corporate or personal details for use in legal or other unintended contexts. Meta has not detailed enforcement mechanisms for non-compliance, though reports suggest refusal could complicate employment amid ongoing restructurings. As AI training relies increasingly on real-world human behavior, this episode underscores growing tensions between tech giants' innovation drives and employee rights in an era of unchecked workplace monitoring.[web:3]