Meta Suspends Employee Keystroke Tracking After Internal Revolt

Meta suspended a controversial program to collect employee keystroke and mouse movement data for AI training after widespread internal backlash and a security breach exposed the collected information across the company.

The Model Capability Initiative (MCI) originated within Meta's elite AI research unit, TBD, this spring as the company sought an edge in the race to automate white-collar work. The program aimed to capture how employees interact with computers—including keystrokes, mouse movements, and screen activity—to train AI models to replicate human computer use for tasks like coding and accounting.

Employee resistance forced concessions

Meta announced MCI as mandatory in April, immediately triggering intense opposition from the workforce. Over 1,800 employees signed a petition against the program, with workers expressing concerns about capturing sensitive personal data including passwords, immigration documents, and financial information.

The resistance took creative forms: employees posted anti-AI memes in internal channels, dubbed the program's screen-recording icon "the Eye of Sauron," and physically spelled out links to the anti-MCI petition using letters on Meta's Burlingame office building. Activists distributed flyers citing National Labor Relations Act protections for workplace organizing.

Meta's chief technology officer Andrew Bosworth initially defended the program's "strict privacy guardrails" and stated there would be no opt-out option. In June, Meta made a partial concession by allowing employees to pause tracking for up to 30 minutes at a time—a move activists called "insufficient."

Security breach ended the program

In late June, an engineer discovered that MCI data had been viewable across the entire company, including full private conversations. The breach contradicted Bosworth's earlier assurances about "tightly controlled data."

Meta suspended the program within hours of the discovery. The company told Business Insider the program would become opt-in if reintroduced, though it remains suspended two months later. Meta declined further comment.

According to internal discussions, the collected data had not been used to train Meta's latest AI models as of July. Bosworth reportedly said the data proved more useful than expected, though Meta has not specified how it was utilized.

Why it matters

The MCI episode illustrates growing tension between companies' hunger for training data and employee privacy concerns as AI development accelerates. Corporate data offers valuable real-world context for AI systems, but Meta's experience demonstrates that mandatory employee data collection can trigger organized resistance even at non-unionized tech companies. The outcome may influence how other firms approach similar initiatives—advocacy firm The Worker Agency called it a "historic win" that will make companies "think twice" about forcing workers to surrender data. The incident also highlights ongoing security challenges as companies rapidly deploy internal AI programs.

Business Insider first reported these details based on interviews with current and former Meta employees and internal company materials.