This article first appeared on GuruFocus.
Meta (META, Financials) is testing a new internal tool that captures employee keystrokes and on-screen activity to help train its artificial intelligence models, according to reports.
The Model Capability Initiative is the name of the system that will collect real-world data on how people use software, such as mouse movements, clicks, and navigation patterns. The information is being gathered from a variety of sources, such as Google and Microsoft services, as well as business applications like GitHub and Slack.
Meta said the goal is to make AI agents better at doing ordinary computer tasks. The company says that watching how real users act is important for making these tools work better and be more useful.
But the endeavor has caused some disagreement within the company. Some workers have said they are worried about privacy threats and the possibility of sensitive information, including personal data or business secrets, getting out.
The program is a response to more general pressure on Meta to catch up in generative AI, where rivals like OpenAI and Anthropic have made progress. It also shows how the rapid growth of AI is making it harder to manage data.
