Meta has announced that it has found a new source of training data for its artificial intelligence models. As part of its efforts to make AI more capable and efficient, the company plans to use data derived from employees' mouse movements and keyboard presses. This information, first reported by Reuters, shows how far technology companies are going in their search for new training sources for AI models.
Redaksiya reports that in response to TechCrunch's request for comment, a Meta representative provided the following statement: “If we are building agents to help people perform everyday tasks using computers, our models need examples of how people actually use them — for example, mouse movements, button clicks, and navigation in dropdown menus. To help, we are introducing an internal tool that will capture such inputs in certain applications to help train our models. Security mechanisms are in place to protect confidential content, and the data is not used for any other purpose.”
This trend reveals a concerning aspect of the AI industry regarding privacy. Last week, it was reported that older startups had their corporate communication archives (e.g., Slack archives and Jira tickets) “scraped” and subsequently converted into AI training data.