Now Meta will track what employees do on their computers to train its AI agents
The internal tool records mouse movements, keystrokes, and screenshots from US-based employees' work computers.
Meta is launching a controversial internal data collection program called the Model Capability Initiative (MCI) to train its AI agents. The tool, installed on US-based employees' work computers, operates within work-related applications and websites, capturing granular interaction data including mouse movements, keystrokes, clicks, and periodic screenshots. According to Meta CTO Andrew Bosworth, this data is critical for the company's Agent Transformation Accelerator (ATA) project, which aims to build AI agents that can perform work tasks. A Meta spokesperson stated the goal is to teach models "how people actually use" computers, with safeguards to protect sensitive content.
The initiative has sparked significant internal backlash, with employees expressing discomfort and asking how to opt out. In response to one such query, Bosworth explicitly stated, "There is no option to opt out of this on your work provided laptop." While Meta claims the data is solely for training AI models and not for performance assessment, the mandatory, pervasive nature of the monitoring raises immediate privacy and ethical concerns. This move highlights the intense, data-hungry race among tech giants to develop capable AI agents, even if it means turning their own workforce into a live training dataset.
- Meta's 'Model Capability Initiative' (MCI) tool records employee mouse movements, keystrokes, clicks, and screenshots on work computers.
- The collected data feeds the 'Agent Transformation Accelerator' (ATA) to train AI models to automate tasks by mimicking human-computer interaction.
- The program is mandatory for US employees on company laptops, with no opt-out, and has caused significant internal backlash according to reports.
Why It Matters
It sets a precedent for workplace surveillance in the AI era, using employee behavior as a core training resource for automation.