Meta paused its internal AI training program after employee keystroke and conversation data was leaked company-wide. CTO Andrew Bosworth said a researcher placed the data in an unauthorized location. The incident adds to Meta's history of privacy scandals as it races to develop advanced AI systems.
Meta's internal AI training program tracked employee keystrokes and private conversations before leaking the data company-wide, Chief Technology Officer Andrew Bosworth said, forcing the company to pause the initiative.
"An internal researcher put the data in a place it wasn't supposed to go," Bosworth, Meta's chief technology officer, said in a statement on Wednesday.
The Model Capability Initiative, Meta's internal AI training program, collected employee keystrokes and private messages to train its large language models. The data was inadvertently exposed across Meta's internal systems, prompting an immediate suspension of the training program, Bosworth said.
The incident raises fresh questions about Meta's data governance as it races to develop advanced AI systems. The company has faced multiple privacy scandals, including the Cambridge Analytica affair and ongoing lawsuits over its AI glasses' recording capabilities, which have drawn scrutiny from regulators in the US and Europe.
Meta's privacy practices have come under increasing pressure as it expands its AI ambitions. The company's privacy policy allows it to use any image shared with Meta AI for training purposes, and it has explored continuously recording audio through its AI glasses prototypes, according to the Financial Times.
The Menlo Park, California-based company has invested heavily in AI infrastructure, with capital expenditures reaching $37 billion in 2025, much of it directed toward building the computing power needed for large language model training. The data governance lapse comes as Meta competes with OpenAI, Google and Anthropic for AI talent and market share.
Apple has declined to partner with Meta over privacy concerns, people familiar with the matter have said, underscoring the reputational damage from repeated data handling incidents. Meta is also facing multiple investigations and lawsuits over privacy violations related to its AI glasses, including a case involving Kenyan workers who alleged they were exposed to graphic content while training Meta's AI.
The company has attempted to address privacy concerns with its hardware products. Meta announced an update to its AI glasses that will disable the camera if the LED recording indicator has been tampered with, after users were found using tape to cover the light. Despite this safeguard, Meta is reportedly testing a prototype that would continuously collect audio while taking photos every few seconds.
Meta shares fell 1.8% in after-hours trading following the disclosure, according to data from the New York Stock Exchange. The stock had gained 34% year-to-date through Wednesday's close, driven by optimism around its AI investments. The data leak could slow Meta's AI training timeline and increase regulatory risk, potentially weighing on investor sentiment toward the stock.
This article is for informational purposes only and does not constitute investment advice.
