Meta installs computer monitoring tool on US employee machines to train AI agents

Meta has begun deploying a monitoring tool across its US workforce called the Model Capability Initiative, which records detailed activity from work computers to feed training data for the company's AI agents. The system captures mouse movements, keyboard input, and intermittent screenshots from within job-related applications and websites.

The company framed the initiative as essential for building effective AI agents. Meta spokesperson Tracy Clayton stated the tool captures examples of how users interact with software—button clicks, menu navigation, task completion patterns—to help the company's models replicate human computer use. The company said it includes safeguards to shield sensitive information and stated the data will not be used for employee performance evaluation.

The tool's deployment was formally announced by Andrew Bosworth, Meta's chief technology officer, in an internal memo that also outlined broader plans for an Agent Transformation Accelerator program. Bosworth described a vision where AI agents would handle primary work functions while employees focus on oversight and improvement.

The initiative has triggered visible pushback from employees. Internal communications revealed staff members questioning the surveillance tool and requesting opt-out mechanisms. When asked directly, Bosworth stated no opt-out mechanism exists for those using company-provided computers, a response that appears to have intensified internal friction over the program.