Meta Starts Collecting Employees' Mouse Movements and Keystrokes for AI Training
Key point
Meta collects input and screen data from employee PCs to use in AI training.
Details
Meta has begun installing the Model Capability Initiative (MCI) on U.S. employees' computers to track mouse movements, clicks, and keystrokes, and has also started capturing periodic snapshots of screens showing work apps and websites.
The goal is to gather training data on human computer interactions that AI still doesn't imitate well, such as selecting dropdown menus and using keyboard shortcuts. An internal memo stated that "every Meta employee can contribute to improving the models just by doing their everyday work."
CTO Andrew Bosworth rebranded the internal AI transition program as the Agent Transformation Accelerator (ATA), presenting a direction where agents mainly perform tasks while humans instruct, review, and refine them. A Meta spokesperson confirmed that MCI data is one of the inputs for model training.
The company explained that this data is not used for performance evaluations, but did not specify exactly what "sensitive content" is excluded. At the same time, Meta began a 10% global workforce reduction starting May 20 and is actively encouraging the use of AI agents for tasks including coding.
Legal scholars pointed out that this approach applies constant surveillance even to white-collar employees, and that it is particularly likely to violate GDPR in Europe. In the U.S., weaker federal restrictions on workplace surveillance make the regulatory gap even more apparent.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.