Meta will record the mouse movements of its employees to train the artificial intelligence

This story, reported by TechCrunch, shows the extent to which technology companies are willing and searching for new sources of training data, which are a key resource for AI models and help them to perform tasks and respond to user queries more efficiently.

New internal tracking tool

Meta has launched a new internal program called the “Model Capability Initiative” (MCI), which will monitor and record the keystrokes, mouse movements and screen content of its employees in the United States. The tool is installed on the official computers of employees and collects data from business applications and websites. According to Meta, the main goal is to improve the ability of their AI models to mimic human interaction with computers, especially in areas where they are currently weak, such as navigating drop-down menus and using keyboard shortcuts.

This strategy is part of a broader plan to develop AI agents that could perform work tasks independently. In a statement to TechCrunch, a Meta spokesperson confirmed the initiative.

“If we’re building agents that help people do everyday tasks with computers, our models need real-world examples of how people actually use them—things like mouse movements, button clicks, and menu navigation. To help, we’re running an internal tool that will record these types of inputs on specific applications to help us with training our models. There are safeguards in place to protect sensitive content, and the data is not used for any other purpose.”

Employees worried, shutting down is not an option

The announcement caused a strong negative reaction from many Meta employees, who expressed discomfort and privacy concerns on internal forums. One of the most popular comments on an internal announcement about the program read: “How can we opt out?”. In response, Meta’s Chief Technology Officer, Andrew Bosworth, stated that there is no option to turn off tracking on official laptops.

The company has assured its employees that the collected data will not be used to evaluate work performance and that there are safeguards in place for sensitive information. However, they did not specify what is considered “sensitive content” that would be exempt from collection. In a subsequent internal memo, with which they tried to reassure the employees, they stated that the content of the screen would be masked during training and that access to raw data would be restricted. The memo also advised employees who are not satisfied with the controls to avoid doing private work on official computers.

Fear of employees training their own replacements

The move comes as Meta is reportedly preparing for another round of layoffs, leading to fears that employees are actually training their own replacements. The initiative is part of the company’s broader focus on artificial intelligence, with chief technology officer Andrew Bosworth laying out a vision where AI agents will “primarily do the work” while being guided and enhanced by human employees.

While the use of internal data to train AI is not unique to Meta, the mandatory nature of this initiative has brought renewed attention to issues of workplace surveillance and data privacy. Legal experts note that such a program could face significant obstacles under the European GDPR regulation, which is much stricter in terms of monitoring employees. For now, the program is limited to Meta’s employees in the USA. This trend also reveals problematic implications for privacy in the AI ​​industry, where yesterday’s internal corporate communications are increasingly becoming the raw material for the new corporate supply chain. Last week, it was announced that legacy startup companies are being searched for their corporate communications (from Slack archives, Jira tickets, and other internal messaging platforms), which could be turned into AI fuel.