Meta is installing tracking software on the work computers of its United States-based employees to capture mouse movements, clicks, keystrokes, and occasional screenshots, using the data to train artificial intelligence (AI) models capable of performing computer tasks autonomously.
The tool was disclosed in a memo to staff in a channel belonging to the Meta Superintelligence Labs team, and will run on a designated list of work applications and websites. The memo framed the effort as a way for employees to improve company models in areas where they still struggle to emulate basic computer-use behaviours, such as navigating dropdown menus and using keyboard shortcuts, telling staff they can help simply by doing their daily work.
Meta spokesperson Andy Stone said the data collected would not be used for performance assessments or any other purpose besides model training, and that safeguards were in place to protect sensitive content. Stone added: “If we’re building agents to help people complete everyday tasks using computers, our models need real examples of how people actually use them.”
The data-gathering announcement came a day after Meta Chief Technology Officer Andrew Bosworth told employees in a separate memo that the company would step up internal data collection as part of its broader Agent Transformation Accelerator initiative, saying Meta would be rigorous about building up data and evaluations for all the types of interactions employees have in their daily work.
The story, first reported by Reuters, shows the lengths to which technology companies are going to find new sources of training data, the lifeblood of AI models that helps programmes learn how to more effectively carry out tasks and respond to user queries.
In January, OpenAI was reported to be asking third-party contractors to upload samples of real work products from previous jobs, including presentations and spreadsheets, with instructions to remove confidential material before submission. The developments reflect a wider industry shift toward harvesting behavioural and workplace data as conventional training datasets become harder to source.
The move has drawn scrutiny over privacy and workplace surveillance, with critics questioning whether assurances against non-training use of the data are enforceable in the long term.
