Meta Pauses Employee Tracking Tool
The AI training program captured staff mouse movements, clicks and keystrokes before a data exposure raised fresh privacy concerns.
Topics
Instagram and WhatsApp parent Meta has paused an internal AI training program that tracked employees’ mouse movements, clicks and keystrokes after sensitive staff data was exposed inside the company.
The program, called the Model Capability Initiative, or MCI, was rolled out in April for US-based employees. Meta said the tool was meant to help train AI models by showing how staff use work-related apps and websites.
The pause was first reported by Business Insider. Reuters also reviewed internal documents showing that data collected through the program had been accessible to Meta employees.
Meta spokesperson Tracy Clayton said the company had “carefully designed this program with privacy safeguards” and had no indication that staff data was improperly accessed.
The company is pausing the program while it investigates, Clayton said.
Reuters reported that MCI was still recording as of Monday afternoon, citing a person familiar with the matter. Meta said the pause was being rolled out and would take time to reach everyone.
The exposed data included full prompts and transcriptions, private conversations, people and performance data, and sensitivity ratings, according to internal documentation reviewed by Reuters.
Business Insider reported that the incident was classified as a SEV 2 security event on Meta’s internal scale, where ‘0’ is the most severe.
The program had already drawn concern inside Meta.
WIRED reported that an internal security notice said employee data across 45,000 Hive tables had been exposed.
Meta Chief Technology Officer Andrew Bosworth told employees in an internal post that the implementation had fallen short of the standards set out in Meta’s privacy review, WIRED reported.
The episode adds to a broader debate over how far companies should go in using employee activity to train AI systems. For Meta, the immediate issue is narrower but more damaging: a tool built to teach AI how workers behave appears to have revived the oldest problem in workplace technology, which is whether workers can trust the system watching them, analysts said.
