Meta Tracks Employee Keystrokes on Google, LinkedIn and Wikipedia for AI Training – What Other Sites Are Being Monitored?

Meta Platforms is reportedly tracking employee keystrokes, mouse movements and on-screen activity through an internal artificial intelligence training programme designed to improve its next generation of AI systems, according to multiple reports.

The initiative, which applies to US-based employees using company-issued devices, has drawn scrutiny over workplace surveillance and how far tech firms are going to train AI agents.

Reports published this week suggest the system captures general digital behaviour across work environments, rather than targeting specific websites such as Google, LinkedIn or Wikipedia.

AI Programme Captures Digital Behaviour

The initiative, internally referred to as the Model Capability Initiative (MCI), is designed to gather real-world interaction data from employees as they work. The system records keystrokes, mouse movements, click patterns and occasional screen snapshots while staff use work-related tools and web-based platforms.

According to CNBC, Meta's internal AI training initiative involves tracking employee activity across commonly used platforms such as Google, LinkedIn and Wikipedia as part of efforts to improve its artificial intelligence systems.

The data is collected from company-issued laptops and is intended to help train AI systems to understand how humans interact with digital interfaces in real working conditions. This includes how users navigate software, switch between tasks and engage with productivity tools.

Meta's objective is to improve AI agents capable of performing complex workplace functions, such as automating administrative processes and interacting with enterprise software in a human-like manner.

Google, LinkedIn and Wikipedia Claims Clarified

Online discussions have linked the monitoring programme to specific websites including Google, LinkedIn and Wikipedia. However, current reporting does not confirm that these platforms are being individually targeted or singled out.

Instead, these sites are frequently used as examples of common workplace browsing activity. Employees may access search engines, professional networking platforms or reference sites during daily tasks, and such interactions could be indirectly captured if they occur on monitored work devices.

The system is not understood to operate as a website blacklist or targeted surveillance tool. Rather, it records broader behavioural signals such as typing and navigation patterns across digital environments used for work.

Keystroke and Mouse Data Used for AI Development

The collected data feeds into Meta's wider strategy to develop AI agents capable of replicating human digital behaviour. These systems are being trained to understand how users complete tasks such as filling forms, navigating menus and interacting with complex interfaces.

Unlike traditional AI training methods that rely heavily on publicly available datasets, Meta's approach focuses on behavioural data generated in real-time work settings. This shift reflects a growing industry trend towards using proprietary workplace data to improve AI performance in enterprise environments.

The Model Capability Initiative is part of a broader effort to advance automation tools that can assist or replace repetitive digital tasks.

Workplace Surveillance Concerns

The rollout of keystroke and screen activity monitoring has raised concerns about employee privacy and workplace surveillance. Reports suggest that participation in the programme is mandatory for employees using company-issued devices, with no opt-out option provided.

Critics argue that even if the data is restricted to work environments, keystroke logging and screen capture techniques raise questions about how much employee behaviour is being observed and analysed.

Key concerns include data transparency, retention policies and whether such monitoring could expand beyond its current scope over time. There is also debate about how clearly employees are informed about what is being collected and how it is used in AI model training.

Growing Use of Behavioural Data in AI Training

Meta's initiative reflects a wider shift across the technology sector towards using behavioural data to train artificial intelligence systems. Companies are increasingly exploring ways to capture how users interact with software in real-world environments, rather than relying solely on scraped internet data.

This includes analysing navigation patterns, workflow behaviour and digital task execution within enterprise systems. Industry analysts suggest this trend could redefine how AI tools are built, particularly in workplace automation and productivity software.

As AI systems become more integrated into daily work processes, the collection of behavioural data is expected to remain a central point of discussion among technology companies, regulators and employees alike.