Meta Briefly Exposed Employee Keystroke Data From AI-Training Program

Meta discovered a significant security misconfiguration that left keystroke logs, screen captures, and conversation transcripts from its employee-surveillance program exposed to all internal staff. The data collection effort, known as the Model Capability Initiative, had been tracking US workers' laptops since April to feed AI training models with human behavior patterns.

The breach affected roughly 45,000 data tables and exposed what Meta's own security team described as "full prompts and transcriptions, private conversations, people and performance data." When the issue surfaced, the company paused the program indefinitely rather than simply patching the access controls.

"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 spokesperson Tracy Clayton told WIRED.

Employee Backlash#

The security failure vindicated long-standing concerns among Meta staff. Over 1,600 workers signed an internal petition last month warning that the tracking program introduced "both security and regulatory risks." Some engineers compared the screen-scraping to exploitation and an invasion of privacy—arguments company leadership had dismissed as necessary sacrifices to train smarter AI systems.

In response to employee questions, Meta's CTO Andrew Bosworth acknowledged the implementation "fell short of the standards outlined in its privacy review," citing misconfigured access control lists as the root cause. Days earlier, Bosworth had assured worried staff that the tracking system used the same protections as other sensitive datasets—a claim the breach undermined.

Ongoing Scrutiny#

Meta operates under a Federal Trade Commission consent decree (through 2040) requiring it maintain robust safeguards against breaches. The company has faced heightened skepticism from employees already demoralized by mass layoffs, a chaotic reorganization, and an intensifying focus on AI projects described internally as "soul-crushing." The security lapse is likely to worsen morale further, even as CTO Bosworth promised greater transparency about the failed system.

Read the full story at WIRED.