How Meta tried, and failed, to grab its employees' data for AI

Meta's elite AI research unit devised a plan to train its chatbots to use computers by capturing employees' screens without explicit consent, according to Business Insider. The initiative sparked an internal revolt after details leaked, ultimately forcing the company to suspend the program.

The program was designed to generate training data by recording employee screens as they performed routine tasks on their computers. The recordings would be used to teach Meta's AI models to interact with software interfaces — a capability known as computer use or agentic AI. The data collection was meant to happen automatically, with employees potentially unaware the footage was being captured for AI training purposes, per Business Insider. Once word of the effort spread internally, employees raised ethical and privacy concerns, and the backlash forced Meta to shelve the initiative.

The move reflects broader industry tension over how tech companies source data for large language models and AI training. As generative AI has become more central to competition, companies have sought ways to expand training datasets — often by mining employee work, customer interactions, or other proprietary information. Meta's effort to harvest employee screen data sits at the intersection of AI ambition and employee privacy, raising questions about consent and what companies can reasonably ask of their workforce in the name of AI development.

The suspension signals that even within Meta, where founder Mark Zuckerberg has positioned AI as central to the company's future, there are limits to how far the organization will push data collection from its own staff. The incident underscores mounting internal scrutiny of AI practices and suggests that large tech workforces are increasingly willing to push back on data practices they view as overreach — a dynamic that may reshape how companies approach AI training in the months ahead.