Just two months after Meta announced plans to track employee keystrokes and mouse movements to train its AI models, the company said it would pause the program after an alleged internal leak.
According to a screenshot obtained by Business Insider, employees’ private conversations, performance data, and transcriptions were accessible across the company. On a scale of 0 to 5—0 being the most severe—the incident was classified as a SEV (severity) 2. Meta confirmed the incident to Business Insider.
According to documents reviewed by Wired, a security notice said that “employee data across 45,000 hive tables”—a data storage format—were exposed, which included employee activity such as “full prompts and transcriptions, private conversations, people, and performance data.”
In response to Fast Company’s request for comment, a Meta spokesperson said: “We have carefully designed this program with privacy safeguards, and while we have no indication at this time that any data was improperly accessed by Meta employees, we’re pausing it while we investigate.”
Meta employees had strong reactions to the internal AI training model, called the Model Capability Initiative, or MCI. Staffers could not opt out of being tracked by the software on company laptops, which gave rise to privacy concerns. Following this tension—and 10% of Meta’s workforce being laid off—CTO Andrew Bosworth said that company morale was near “the worst it’s ever been.”
In May, Meta told Fast Company that its “[AI] models need real examples of how people actually use [computers]—things like mouse movements, clicking buttons, and navigating drop-down menus.” A spokesperson for the company also added that there were “safeguards in place to protect sensitive content.”
Still, employees created an online petition protesting the use of the software. They posted flyers across Meta offices calling on their colleagues to sign it.
“When employees asked what privacy reviews were conducted, including any ‘people data reviews’ (which are required for processing employee data), no completed privacy reviews were provided,” read the petition, which had 1,600 signatures as of June 3. “The outlined privacy mitigations were vague, and leadership’s confidence in them appeared limited—evidenced by the selective opt-out afforded to executives.”
