Meta Employees Protest Mouse-Tracking Tool Built to Train AI Models

Meta employees at multiple U.S. offices distributed protest flyers. The action targets an internal tool that records mouse movements and keystrokes to train Meta’s AI models.

The flyers appeared in meeting rooms, on vending machines, and atop toilet paper dispensers. They direct staff to a petition demanding Meta withdraw the tool, called the Model Capability Initiative.

Mouse-Tracking Tool Collects Data Across Hundreds of Sites

Meta installed the Model Capability Initiative on U.S. employees’ work computers. The software captures clicks, keystrokes, and periodic screenshots. It runs across hundreds of applications, including Google, LinkedIn, and Wikipedia.

Meta says the data trains AI agents to mimic real human behavior on the web. According to a spokesperson, the models need real examples of how people use software. The company cited button clicks and dropdown menus as key inputs.

Several Meta employees called the program “dystopian” in interviews last month. Workers fear the tool could expose passwords and unreleased product details. They also flagged concerns about personal information, including immigration status, health records, and family details.

Petition Lands Days Before 8,000-Job Layoff

The petition builds on internal opposition that has grown since the April rollout. The flyer campaign on May 12 marked the first coordinated public action by U.S. staff against the program.

Pressure on Meta’s workforce is compounding. Chief people officer Janelle Gale said in late April that the company will cut 8,000 roles on May 20. The move is part of an efficiency drive tied to AI spending. Another 6,000 open positions will go unfilled.

Across the Atlantic, Meta employees in the United Kingdom announced a separate organizing effort this week. They are partnering with the United Tech and Allied Workers union.

Privacy Trade-Off Sits at AI Training Frontier

Meta says safeguards prevent the capture of certain sensitive content. However, it has not detailed the technical scope of those filters. Affected employees argue that consent terms are coercive, given the impending layoffs.

The dispute highlights a wider problem for large AI developers. Quality human behavioral data is increasingly scarce, and companies are turning to their own workforces to fill the gap.

Whether the May 12 protest pressures Meta to alter the program remains unclear. The outcome may signal how much leverage employees retain in an AI-driven cost-cutting cycle.