Meta says it is rolling out internal software that will collect activity data from some U.S. employees' work computers to help train AI systems that can operate software on their own.

What happened

Reuters reported that internal memos described a tool for Meta staff that captures mouse movements, clicks, keystrokes, and periodic screenshots on a designated set of work apps and websites. TechCrunch and Fortune both published Meta's response, which framed the effort as a way to give its models real examples of how people use computers for everyday tasks.

Meta spokesperson Andy Stone said the company needs examples of actions such as clicking buttons, navigating dropdowns, and using common interface patterns if it wants to build agents that can complete computer-based work. Meta also said the collected data is meant for model training, not employee evaluation, and that safeguards are in place to protect sensitive content.

Why it matters

The immediate news is not a new chatbot release. It is a clearer view into how frontier AI labs are trying to gather training data for computer-use agents, an area where synthetic data and public web text do not fully capture real workplace behavior.

The more conservative takeaway is that Meta appears willing to use internal employee workflows as a proprietary data source for agent training. That could give its models better examples of how people actually move through enterprise software, while also raising obvious questions about workplace surveillance, consent, and how far companies will go to feed the next generation of AI systems.