Breaking

Meta Pauses AI Program That Trained on Staff PC Activity After Internal Data Exposure

June 23, 2026 at 23:52 EDT

Security
Policy & Regulation
Foundation Models

Meta has paused an internal program called the Model Capability Initiative (MCI), which collected employees' keystrokes, mouse movements and other computer activity to train AI models, after the collected data became unintentionally accessible across the company due to an internal security issue. The pause took effect around June 22, 2026, with Meta saying it would halt the program until its investigation is complete.1

June 22, 2026 · Meta

Meta Pauses Employee-Tracking AI Program After Data Leak

A permissions misconfiguration in an internal database left keystroke, mouse and activity data — collected to train "human-like" AI — accessible far beyond its intended scope, forcing Meta to suspend the "Model Capability Initiative" (MCI).

~45,000

Data tables in the affected database

4 hrs

Until a patch was applied — but it did not fully resolve the issue

~2 mo

From April 2026 launch to June pause

Timeline of the SEV

APR 2026

MCI launches for US staff; no opt-out

→

JUN 18

Permissions error found; patch fails to fully fix

→

JUN 22

Program paused pending data-control verification

What MCI captured from employees' computers

Everyday workstation activity, harvested as training data for agentic AI.

Keystrokes

Mouse moves

Click positions

Screen content

Meta's position

"Carefully designed privacy protections" with "no indication of improper data access." Says it has "collected enough data to evaluate the tool's long-term value."

Staff concerns

Privacy and surveillance-culture worries from the start. Workers say that when risks were raised, "leadership doubled down rather than acknowledging" them.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.