Meta has identified a new source for training its AI models: its own employees. The company intends to utilize data from its staff’s mouse movements and keystrokes to develop more advanced and efficient artificial intelligence.
First reported by Reuters, this highlights the lengths tech companies are willing to go to acquire new training data sources—crucial for AI models to improve task execution and user response.
A Meta spokesperson told TechCrunch:
“If we’re creating agents to assist with everyday computer tasks, our models require real examples of actual usage—such as mouse movements, button clicks, and menu navigation. To support this, we are introducing an internal tool to capture these inputs on specific applications for model training. Safeguards are in place to protect sensitive content, and the data is not used for other purposes.”
This trend underscores the AI industry’s privacy concerns, as internal communications are increasingly repurposed for corporate supply chains. It was reported last week that startups’ corporate communications are being repurposed as AI training data, including Slack archives, Jira tickets, and other internal messages.