Meta Plans to Harvest Employee Keystrokes and Mouse Clicks to Train AI Agents
Meta is preparing to deploy new monitoring software on the work devices of its US employees that will collect granular behavioural data -- including mouse movements, clicks, and keystrokes -- for the purpose of training AI agents to perform real-world tasks, according to reports published on April 22, 2026. The initiative represents one of the most extensive uses of employee data for AI training purposes disclosed by a major technology company.
Background
The development of AI agents -- software systems capable of autonomously performing complex, multi-step tasks -- has become a central focus of the technology industry. Unlike large language models that primarily generate text, AI agents are designed to take actions: browsing the web, writing and executing code, managing files, and interacting with software applications. Training such agents requires vast quantities of data showing how humans actually perform these tasks in real-world environments.
Meta has been investing heavily in AI agent development as part of its broader artificial intelligence strategy, which includes the open-source Llama model family and a range of AI-powered features across its social media platforms. The company has positioned AI as central to its long-term growth strategy.
Key Developments
According to reports, Meta plans to install software on employees' work computers that will capture detailed records of their interactions with their devices, including every mouse movement, click, and keystroke. This data will be used to create training datasets for AI agents, teaching them to replicate the workflows and decision-making patterns of human workers. The data collection is expected to begin with US-based employees before potentially expanding to other regions.
The scope of the data collection is notably broad. Unlike anonymised usage statistics or aggregate behavioural data, the planned collection would capture the specific sequence of actions taken by individual employees as they perform their jobs, potentially including sensitive information about internal processes, communications, and proprietary workflows.
Why It Matters
The initiative raises significant questions about employee privacy, consent, and the appropriate use of workplace data. While employers generally have broad legal authority to monitor activity on company-owned devices, the use of such data to train commercial AI products -- which could ultimately be sold to third parties or used to automate the very jobs being monitored -- introduces ethical dimensions that go beyond traditional workplace surveillance.
Privacy advocates and labour rights organisations are expected to scrutinise the programme closely. The collection of keystroke and mouse data is particularly sensitive because it can reveal not just what employees do, but how they think and work, potentially exposing proprietary knowledge and individual working styles. The initiative also comes at a time of heightened public concern about AI's impact on employment, with many workers worried that AI tools trained on their own labour could eventually be used to replace them.
What's Next
Meta has not publicly confirmed the full details of the programme, and it is unclear whether employees will be given the opportunity to opt out of data collection. Labour lawyers and privacy experts are expected to weigh in on the legality of the initiative under existing US employment and privacy law. The programme may also attract attention from state regulators, particularly in California, which has some of the strongest employee privacy protections in the country.
Sources: TechStartups; MIT Technology Review
