Meta Deploys Monitoring Software in U.S. Offices for AI Development

Meta has introduced new monitoring software on computers of its U.S.-based employees, capturing every mouse movement, click, and keystroke. This initiative aims to enhance artificial intelligence models by training them to perform tasks typically requiring human interaction, such as selecting dropdown menus and using keyboard shortcuts. The program, named Model Capability Initiative (MCI), also takes occasional screen snapshots for context, with all data directed towards improving AI capabilities.

Internal memos obtained by Reuters reveal the explicit purpose of MCI: to refine AI in areas where it currently falters, particularly human-computer interaction. A staff AI research scientist highlighted that employees can contribute to AI improvement merely by performing their daily tasks. This move follows a recent announcement by Meta CTO Andrew Bosworth, who emphasized increased internal data collection for the company's AI efforts, rebranded as Agent Transformation Accelerator. Bosworth envisions a future where AI agents handle most work, with humans overseeing and refining their operations through a "closed loop" learning system.

The deployment of MCI is part of Meta's broader AI-first strategy under CEO Mark Zuckerberg. While the company undergoes layoffs, shifting engineers into an Applied AI team focused on developing autonomous AI agents to manage coding, testing, and product shipping, this initiative underscores Meta's commitment to integrating AI deeply into its operations. The spokesperson assured that the surveillance is not for performance evaluations but for enhancing AI models, with safeguards in place to protect sensitive content.