Meta has temporarily halted a controversial employee monitoring program after sensitive data was inadvertently shared with the entire company.

The program, known as the Model Capability Initiative, had been quietly gathering keystrokes, mouse movements, and screen content from employee laptops in the US since April.

Surveillance effort draws internal backlash

Meta introduced the initiative to help train AI systems on how people interact with software, claiming that employee activity provided superior training data compared to external contractors. Over 1,600 staff members signed a petition opposing the effort, citing risks to security and compliance. An engineer highlighted the lack of consent for screen capture as a privacy violation.

CEO Mark Zuckerberg supported the program in a recorded discussion, noting that AI models benefit most from observing skilled individuals at work and that Meta employees offered better examples than typical contractors.

Internal leak prompts immediate pause

A security alert disclosed that information from 45,000 database tables had been made accessible company-wide. This included private chats, prompts, transcripts, and performance metrics. Spokesperson Tracy Clayton stated there is no indication of unauthorized access but confirmed the program is on hold during the review process. Employees voiced strong concerns on internal channels, with one former worker describing the situation as a predictable outcome.

CTO Andrew Bosworth admitted internally that the initiative did not meet Meta's privacy standards. This marks another AI-related security issue for the company, following earlier events involving unauthorized AI actions and account vulnerabilities. The pause may ease current tensions, though ongoing workforce challenges from restructuring continue to test employee relations.