Meta employees protest mouse-tracking tool ahead of layoffs

On May 12 Meta employees at multiple U.S. offices distributed flyers calling on the company to withdraw the Model Capability Initiative, a mouse‑tracking tool installed on staff computers that records clicks, keystrokes and periodic screenshots to train artificial intelligence.

The flyers appeared in meeting rooms, on vending machines and on toilet paper dispensers and directed staff to a petition opposing the program. The distribution came days before Meta plans to eliminate 8,000 roles on May 20.

The Model Capability Initiative rolled out in April and runs across hundreds of applications on company machines, including Google, LinkedIn and Wikipedia, according to internal descriptions and employees. Meta describes the data as examples of button clicks and dropdown selections used to train AI agents to mimic human interactions on the web.

Meta provided a statement noting that models need real examples of how people use software and that safeguards are in place to block certain sensitive content. The company has not publicly detailed the technical scope of those filters.

Employees raised concerns that the software can capture sensitive inputs such as passwords, unreleased product details and private information including immigration status and health records. Several workers described the program as “dystopian.” Staff also objected to consent terms they called coercive, arguing that impending layoffs made refusal difficult.

Pressure on the workforce has increased since chief people officer Janelle Gale informed employees in late April that 8,000 roles would be cut and roughly 6,000 open positions would go unfilled as part of an efficiency plan tied to AI spending. Employees said the timing complicates decisions about participation in internal data programs.

Workers in the United Kingdom launched a separate organizing effort this week and are partnering with the United Tech and Allied Workers union. The petition and flyer campaign ask Meta to stop collecting behavioral data from employee work computers and to set clearer limits on what the tool records.

The dispute reflects a wider practice among AI developers of using internal interaction data to train models because high-quality human behavioral data is limited. Company officials have defended the project as necessary for building more capable AI agents, while employees continue to raise privacy and security concerns.