Privacy at work

Meta records mouse movements and keystrokes by employees 'to train AI'

The company will also 'occasionally' record PC screens. In the US, at federal level, these activities are legal

22 April 2026

4' min read

Translated by AI

Versione italiana

Key points

Work reorganisation
Dismissals in the technology sector
Reorganisation of tasks and structures
A system to monitor employees?

4' min read

Translated by AI

Versione italiana

Meta is installing new tracking software on employee computers in the United States to catch mouse movements, clicks and keystrokes, to be used to train its artificial intelligence models. This initiative is part of a larger project to create AI agents capable of performing work tasks autonomously, as communicated by the company to employees in internal memos viewed by the Reuters news agency.

The tool, dubbed the Model Capability Initiative (MCI), will run on corporate apps and websites and will also occasionally capture shots of the content displayed on employees' screens, according to one of the memos, published on Tuesday by an AI researcher at the company in a channel dedicated to the Meta SuperIntelligence Labs team responsible for creating models.

The aim, according to the memo, is to improve the company's AI models in areas where they struggle to replicate human computer interaction, such as selecting from drop-down menus and using keyboard shortcuts.

"This is where all Meta employees can contribute to improving our models simply by doing their daily work," the statement reads.

The company that owns Facebook and Instagram is moving decisively to integrate artificial intelligence into its workflows and reorganise its workforce around this technology, arguing that this will make the company more efficient.

Meta's CTO, Andrew Bosworth, told employees in a separate memo released on Monday that the company will intensify internal data collection as part of these 'AI for Work' efforts, now renamed the Agent Transformation Accelerator (ATA).

"The vision we are aiming for is one where our [AI] agents primarily do the work. Our role will be to guide them, review them and help them improve,' Bosworth said. The goal, he added, is for the agents to 'automatically see where we think they need to intervene, so they can improve next time'.

Bosworth did not explicitly specify how these agents would be trained, but stated that Meta would be 'rigorous' in 'collecting data and evaluations for all types of interactions we have during our work'.

Meta spokesman Andy Stone confirmed that MCI data would be among the inputs.

Work reorganisation

Stone stated that the data collected through MCI would not be used for performance evaluations or any other purpose besides model training, and that security measures had been put in place to protect 'sensitive content', but without specifying which types of data would be excluded from collection.

"If we are creating agents to help people complete everyday tasks using computers, our models need real-world examples of how people actually use them, such as mouse movements, button clicks and navigation in drop-down menus," Stone said.

The push to automate functions previously performed by human staff reflects a general trend among major US companies this year, especially in the technology sector.

Artificial intelligence tools have taken Silicon Valley by storm with their ability to handle complex tasks such as creating apps and organising large volumes of data with limited human supervision, causing the shares of traditional software companies to plummet and prompting some executives to plan massive staff cuts.

Dismissals in the technology sector

Meta plans to lay off 10% of its global workforce as of 20 May and is considering further significant cuts by the end of the year.

Amazon.com also reduced its workforce by 30,000 in recent months, or almost 10 per cent of its employees, while fintech company Block almost halved its workforce in February.

The reorganisation of tasks and structures

Internally, Meta has urged employees to use artificial intelligence agents for programming and other tasks, even if this slows them down in the short term. It also eliminated distinctions between certain job functions in favour of a new generic qualification called 'AI builder'.

Last month, the company created a new team of applied AI (AAI) engineers with the aim of improving the programming capabilities of AI models and using them to create AI agents capable of doing most of the work of developing, testing and releasing Meta's future products and infrastructure.

Earlier this month, Meta started to transfer 'competent' software engineers to the AAI team.

A system to monitor employees?

Logging of activities on computers and screen capture technologies have historically been used by companies to detect misconduct or non-work activities of employees, said Ifeoma Ajunwa, professor of law at Yale University.

The decision to record employees' keystrokes takes data collection targets to the next level, he added, subjecting white-collar workers to real-time surveillance previously only experienced by drivers and on-call workers.

"In the United States, at the federal level, there is no limit to the surveillance of workers," Ajunwa said, adding that state laws require at most that workers be informed in a general way when employers monitor them.

European law would probably prohibit such monitoring, said Valerio De Stefano, a law professor at York University in Toronto and an expert in technology and comparative labour law.

In some countries, such as Italy, the use of electronic monitoring to track employees is explicitly illegal, while in Germany, courts have ruled that employers may only use keystroke logging in exceptional circumstances, such as the suspicion of a serious crime.

Moreover, De Stefano said, this practice would probably be considered a violation of the European General Data Protection Regulation (GDPR).