Meta halts worker tracking for AI training due to privacy fears

Meta has suspended an internal program that tracked employee computer usage after staff raised privacy concerns and the company discovered collected data was potentially accessible to anyone inside the firm, the BBC reported. The program, named the Model Capability Initiative, began two months ago to gather data on how people used computers — including mouse clicks and keystrokes — for training artificial intelligence models [1]. Meta paused it on Monday after realizing some collected data had been left potentially accessible internally [1]. A Meta spokesman confirmed the program is “on pause for now” and said the company had “no indication at this time that any data was improperly accessed by Meta employees” [1]. Employee pushback had been building for weeks. Nearly 2,000 Meta workers signed a petition demanding the program be cancelled [1]. In an initial response, Meta said workers could opt out of tracking for up to 30 minutes at a time [1]. One current employee, speaking anonymously, called that “just an attempt at damage control” [1]. Another said that while many technical staff support improving Meta’s AI models to compete with Anthropic and OpenAI, the tracking “was forced on us, there was no consent” [1]. The tracking controversy unfolded alongside broader internal strain. Meta is spending up to $145bn on AI initiatives this year and has conducted extensive layoffs and reorganizations [1]. “I’ve never seen morale here so bad,” one employee said [1]. A former employee described the direction of the company as “exhausting and depressing” [1]. Generative AI systems learn patterns from training data to produce text, images, and other outputs [2]. The demand for training data has intensified since the AI boom of the 2020s, driven by advances in large language models [2]. Companies across sectors have adopted generative AI, but the technology has also raised concerns about data practices and consent [2]. The Center for Democracy and Technology, a Washington-based nonprofit, advocates for transparency, accountability, and limiting the collection of personal information [4]. Its work focuses on enabling individuals to use the internet while reducing potential harms [4]. Employee monitoring programs that feed AI training pipelines sit at the intersection of workplace privacy and data governance questions the organization has long tracked [4]. Meta’s internal tensions reflect a wider industry dynamic. A 2025 study interviewing 25 leading AI researchers from labs including Google DeepMind, OpenAI, Anthropic, and Meta found that 20 identified automating AI research as one of the most severe and urgent AI risks [7]. Participants predicted AI agents would gradually transition from assistants to autonomous AI developers, though they disagreed on timelines and governance mechanisms [7]. Seventeen of 25 researchers expected advanced AI systems to be increasingly reserved for internal use at companies or governments, unseen by the public [7].