Meta rolled out tracking software on U.S. employees' work computers. It records mouse movements, clicks, keystrokes, and some screenshots. All this data goes into their AI training system.
A memo from the Meta Superintelligence Labs team explained the plan. It targets certain work apps and websites. The goal is to teach models basic computer skills such as handling dropdown menus and keyboard shortcuts. Staff can help just by doing their daily tasks.
“If we’re building agents to help people complete everyday tasks using computers, our models need real examples of how people actually use them,” Meta told Fortune. The company uses details like mouse movements and button clicks. They have safeguards for sensitive content. The data serves training only.
Building Real AI Agents
Meta wants AI agents that perform white-collar jobs. These tools know lots of info but miss examples of actual computer use. They compete hard with OpenAI and Anthropic.
Industry Wide Data Chase
Everyone hunts for solid training data. OpenAI asked contractors via Handshake AI for real files like PowerPoints and spreadsheets. Workers must remove confidential details first.
Meta grabbed a 49 percent stake in Scale AI for more than 14 billion dollars last year. The ex-CEO now heads Meta Superintelligence Labs. Meta plans up to 135 billion dollars in AI spending for 2026. At the same time they prepare to cut as much as 20 percent of staff with layoffs starting in May.
