.png)
META Platforms has introduced new tracking software on employees’ work computers in the United States, capturing detailed user activity including mouse movements, clicks, keyboard input and periodic screen captures, as part of its push to train advanced artificial intelligence systems.
The initiative, known internally as the Model Capability Initiative (MCI), is designed to support the development of AI agents capable of performing complex workplace tasks independently, according to internal memos seen by Reuters.
The system reportedly operates across work-related applications and websites and intermittently records screen content displayed on employees’ devices.
The data is being collected to improve AI models’ ability to replicate human-computer interaction, including tasks such as navigating dropdown menus and using keyboard shortcuts.
A memo posted by a Meta AI research scientist within the company’s Meta SuperIntelligence Labs described the objective as enabling AI systems to learn from real workplace behaviour.
“Here, all Meta employees can help our model improve simply by doing their daily work,” the memo stated.
Meta, the parent company of Facebook and Instagram, has increasingly integrated artificial intelligence into its internal workflows while restructuring parts of its workforce around AI-driven operations, arguing that the shift will improve efficiency.
In a separate internal communication, Chief Technology Officer Andrew Bosworth said the company would expand internal data collection efforts under its “AI for Work” programme, now rebranded as the Agent Transformation Accelerator (ATA).
“The vision we are building is one where our agents do the primary work and our role is to direct, review and help them improve,” Bosworth said.
He added that the aim is for AI agents to automatically identify areas where human intervention is needed so that systems can improve over time.
Bosworth did not provide detailed information on how the models would be trained but said Meta would take a “rigorous” approach to building datasets and evaluations from real workplace interactions.
Meta spokesperson Andy Stone confirmed that data collected through MCI would be used as training input for AI systems.
He stressed that the information would not be used for employee performance evaluation or any purpose other than model training, adding that safeguards had been introduced to protect sensitive content, though he did not specify exclusions.
“If we’re building agents to help people complete everyday tasks on computers, our models need real examples of how people actually use them — things like mouse movements, button clicks and dropdown navigation,” Stone said.
The move reflects a wider trend across major US technology firms accelerating automation of tasks traditionally performed by human workers, as artificial intelligence tools become increasingly capable of handling complex functions such as software development and large-scale data processing with minimal human oversight.
The shift has contributed to pressure on traditional software firms and prompted some executives to plan significant workforce reductions.
Meta itself has announced plans to cut 10 per cent of its global workforce from 20 May and is considering further reductions later in the year. Amazon has also reduced its corporate workforce by around 30,000 employees in recent months, while fintech company Block has cut nearly half its staff.
Internally, Meta has encouraged employees to use AI agents for coding and other tasks, even if this slows output in the short term. The company has also begun consolidating job roles under broader titles such as “AI builder”.
Last month, Meta established a new engineering unit, Applied AI (AAI), aimed at enhancing its AI coding capabilities and developing agents capable of handling large portions of product development, testing and infrastructure deployment. Engineers deemed highly skilled have already been reassigned to the division.
However, the expansion of workplace monitoring has raised concerns among legal and academic experts about the implications for employee privacy.
Ifeoma Ajunwa, a labour law professor at Yale University, said keystroke logging and screen capture technologies were traditionally used to detect misconduct or non-work-related activity, but the latest developments significantly expand the scope of surveillance.
She noted that US federal law imposes no strict limits on workplace monitoring, requiring only general disclosure to employees in some jurisdictions.
In contrast, European legal frameworks are likely to prohibit such practices, according to Valerio De Stefano, a labour law professor at York University in Toronto.
He said that in some countries such as Italy, electronic monitoring of productivity is strictly illegal, while in Germany courts have ruled that keystroke logging may only be used in exceptional cases involving serious criminal suspicion.
De Stefano added that such practices could also violate the European Union’s General Data Protection Regulation (GDPR), warning that increased awareness of surveillance at work could shift workplace power dynamics further in favour of employers. - April 22, 2026