Sources: Meta Accidentally Exposed Employees' MCI Keystroke-Tracking Data to Each Other, Including Full Prompts and Private Conversations Collected for AI Training
Summary
Wired reports, citing sources, that Meta's Model Capability Initiative — the employee monitoring program capturing keystrokes, mouse clicks, and screenshots to train AI agents — accidentally exposed workers' data to each other, including full prompts and private conversations. MCI had already prompted fierce backlash, with 1,600 employees signing a petition before Meta issued a June 2 memo scaling the program back without ending it. The internal data breach converts what had been an ethical controversy into a concrete security incident.
Shared on Bluesky by 2 AI experts
-
Katie Drummond @katie-drummond.bsky.social
NEW: Insane screwup inside Meta. The company exposed worker keystroke data that was being used to train AI — making it potentially accessible to anyone at the company. The data included personnel and performance info, …
View on Bluesky →
Originally reported by wired.com
Read the original article →Original headline: Sources: Meta Accidentally Exposed Employees' MCI Keystroke-Tracking Data to Each Other, Including Full Prompts and Private Conversations Collected for AI Training