Meta has deployed "Model Capability Initiative," a surveillance tool capturing employee keystrokes, mouse movements, and screenshots on work computers to train AI agents. The tool monitors activity in Gmail, GChat, VS Code, and internal apps. Meta joins Anthropic, OpenAI, and Microsoft in pursuing computer-use AI agents trained on real-world user behavior.
Safety
Irony as Meta staff unhappy about running surveillance software on work PCs
Meta's Model Capability Initiative harvests employee keystrokes and screenshots to train AI agents—a practice that has sparked internal backlash while Anthropic, OpenAI, and Microsoft pursue similar surveillance-driven AI training programs.
Wednesday, April 22, 2026 12:00 PM UTC2 MIN READSOURCE: Hacker NewsBY sys://pipeline
Tags
safety
/// RELATED
Infrastructure6d ago
Linux 7.0 Broke PostgreSQL: The Preemption Regression Explained
A 50% PostgreSQL performance cliff on Linux 7.0: AWS engineer traces the regression to the kernel's removal of PREEMPT_NONE and its clash with spinlock-based buffer management.
StrategyApr 22
Now Meta will track what employees do on their computers to train its AI agents
Meta is deploying MCI, a tool that records employee keystrokes, mouse movements, and screenshots to train AI agents for workplace task automation.