Meta Exposed Employee Data From AI Tracking Program
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Meta left data from 45,000 internal hive tables accessible company-wide, with the exposed information believed to include keystrokes, mouseclicks, screen content, full prompts and transcriptions, private conversations, and performance data from US employees.
- Tracy Clayton, a Meta spokesperson, confirmed an investigation is underway, stating the program was designed with privacy safeguards and that 'we have no indication at this time that any data was improperly accessed by Meta employees.'
- CTO Andrew Bosworth told employees the tracking program's implementation 'fallen short of the standards outlined in its privacy review,' and the security incident has been marked as closed.
- More than 1,600 Meta employees signed an internal petition last month protesting the laptop surveillance—known as the Model Capability Initiative—warning of 'security and regulatory risks' and unauthorized disclosure potential.
- Mark Zuckerberg, defending the data-gathering project in leaked audio, said 'AI models learn from watching really smart people do things' and argued Meta employees are significantly more intelligent than contractors who could produce equivalent data.
- Meta recently began offering limited exemptions to the monitoring, allowing staffers to briefly disable surveillance for sensitive tasks like scheduling personal appointments, though some employees continue demanding the tracking be stopped entirely.
- The Model Capability Initiative launched in April as part of a broader AI reorganization that moved roughly 6,500 employees into new roles on an Applied AI team, with some staffers describing their reassigned work as 'menial and soul-crushing.'
Why it matters: A company already struggling with internal morale—after mass layoffs, reorganization, and a contested AI pivot—now faces a self-inflicted security breach that validates the 1,600+ employees who warned the surveillance program was a liability. With keystrokes and private conversations left exposed across 45,000 data tables, Meta's CTO has publicly admitted the program fell short of its own privacy standards, deepening distrust among engineers who were reassigned to AI work they describe as 'soul-crushing.'
Ask SkimNews
