On Tuesday, 21 April, Meta notified US employees that it was installing software on their desktops to record mouse movements, keystrokes, and screenshots.
The programme is called the Model Capability Initiative (MCI). The data it collects goes directly into a pipeline for training the company's AI agents.
Two days later, on Thursday, 23 April, the company announced it was laying off 8,000 employees, or 10% of its global workforce, starting on 20 May.
At the same time, the filling of 6,000 open positions was cancelled. The total reduction amounts to 14,000 jobs.
It was not a coincidence that these two announcements came in the same week. They outline a strategy.
What MCI actually does
MCI operates on a predetermined list of work applications and sites. According to internal messages seen by CNBC, this list includes Google, LinkedIn, Wikipedia, GitHub, Slack, and Atlassian.
Meta's own platforms, Threads and Manus, are also included. The initial version of the list also featured competing AI tools, including ChatGPT and Anthropic's Claude, which were later removed.
According to Meta's statement, the tool does not operate on personal searches or on applications outside the approved list. Screenshots are taken intermittently, not continuously.
Several employees, in internal messages seen by CNBC, described the programme as "dystopian"
The company claims there are safeguards in place for "sensitive content" and that the data is not used for other purposes, but it has not specified what is meant by sensitive content or how those safeguards function in practice.
Several employees, in internal messages seen by CNBC, described the programme as "dystopian." A specific concern is that the tool could capture passwords, health data, information about employees' immigration status, and confidential details about new product development.
CTO Andrew Bosworth confirmed in an internal communication that there is no "work laptop logout option."
Why is Meta doing this?
The reason is pragmatic. Meta is developing AI agents to perform routine office tasks autonomously. For these agents to function, they require data that shows how people actually use computers in a work environment: the order in which menus are used, how forms are completed, and which navigation patterns follow specific tasks.
Such data does not exist on the Internet in a usable form; it exists only in the working habits of real users on real applications.
Meta’s employees provide an equivalent service at no additional cost, with no right of refusal, as a condition of continued employment
Meta has 78,865 employees. Each, working on approved applications, becomes a source of data. This is is explicitly stated in an internal memo seen by Reuters.
An alternative approach used by OpenAI involves asking external collaborators to provide actual business documents, PowerPoint presentations, and spreadsheets from previous jobs, addressing a similar problem but with external actors who are paid for their input.
Meta’s employees provide an equivalent service at no additional cost, with no right of refusal, as a condition of continued employment.
Extent of layoffs in context
In January, Meta eliminated more than 1,000 positions in its Reality Labs division, closing several VR studios. In March, it laid off a further 700 employees across at least five departments: Facebook Social, Recruiting, Sales, Global Operations, and Reality Labs.
According to HR director Janella Gale’s internal memorandum, the May wave is structural rather than disciplinary. The company is reorganising teams into AI-focused work units and introducing new job categories: “AI builder”, “AI pod lead”, and “AI org lead”.
Laid-off employees in the US receive 16 weeks of base pay plus two additional weeks for each year of service, and 18 months of health insurance
From November 2022 to May 2026, Meta eliminated more than 35,000 jobs, including cuts in 2022, 2023, and three earlier rounds in 2026.
Between 2024 and 2025, the company aggressively rehired staff and ended 2025 with 78,865 employees, a 6% increase compared to the previous year. Now, it is cutting more deeply than it ever hired.
Laid-off employees in the US receive 16 weeks of base pay plus two additional weeks for each year of service, and 18 months of health insurance.
Financial background
Meta invested between $115 billion and $135 billion in AI infrastructure in 2026. Last year, it paid over $14 billion for a 49% stake in Scale AI, a data mining company.
Alexandr Wang, former CEO of Scale AI, now leads Meta Superintelligence Labs, the division under which the MCI programme operates.
Shares of Meta fell 2.4% on Thursday when the layoffs were announced, but quickly recovered. Year to date, shares are roughly at the same value as at the beginning of the year.
Legal framework and limitations
The MCI programme, in its current form, does not apply to Meta's operations in the European Union or the United Kingdom.
The General Data Protection Regulation imposes tests of legitimate interest and data minimisation that such a tracking system would not easily satisfy.
The UK Information Commissioner has a separate code of employment practices, which would create additional regulatory risk.
MCI is currently legal under US law, provided the company complies with conditions being defined in several state privacy laws
Cornell researchers have already publicly raised questions about employee consent and compensation for using their work behaviour as commercial training data.
In the United States, where there is no comprehensive federal data privacy law, the company has significantly more flexibility.
MCI is currently legal under US law, provided the company complies with conditions being defined in several state privacy laws.
Wider context in the sector
Meta is not the only company affected. Amazon cut 30,000 corporate jobs in 2026. Block eliminated almost half its staff in February. The common denominator in nearly all these cases is the same argument: AI can take over the displaced workload, making investment in large human teams unprofitable.
What is specific about Meta is not only the scale of the cuts, but also the explicit feedback loop. The company does not replace employees with AI developed independently of them; instead, it uses these employees, in the period leading up to their dismissal, to train the systems that will take over their jobs.
The MCI memo describes this as the contribution employees make "for all kinds of everyday computing tasks." However, the benefit of that contribution belongs entirely to the company. The employees who provide it do not own any part of what they have built.
In internal and public communications over the past year, Zuckerberg has described a future where AI agents take over a significant portion of cognitive work.
In this narrative, the transitional period when employees are still needed is also when they are most useful as a source of data. The MCI programme puts that logic into operational practice.
If the American precedent becomes the standard, European regulators will have to respond – with new rules, new restrictions, or even an outright ban on the implementation of such programmes in Europe
Regulators in Europe and the United Kingdom are monitoring the situation. The first lawsuits in the United States are expected in the coming months. Meta will publish results for the first quarter of 2026 on Wednesday, 29 April.
This will be the first financial report following the announcement of the MCI programme and layoffs, and the first test of whether the market will reward this strategy in the long term.
The more significant issues, however, are not financial. Does MCI actually work as a data collection method, or is the value of that data overestimated?
AI agents intended to replace office workers have yet to demonstrate that they can handle more complex tasks without constant human supervision. Meta is investing $135 billion on the assumption that this will change – an assumption that has yet to be proven in practice.
An equally important question is how many companies will adopt this model. Amazon, Google and Microsoft are monitoring the situation.
If Meta demonstrates that employees can be used as a source of training data without significant legal or reputational consequences in the US, it is logical to expect the same approach to be replicated. The boundary between the workplace and the AI training lab then becomes systemic rather than exceptional.
For European regulators, the situation is even more complex. MCI is currently limited to the US, but Meta operates across the EU, employing tens of thousands of people.
If the American precedent becomes the standard, European regulators will have to respond – with new rules, new restrictions, or even an outright ban on the implementation of such programmes in Europe.
The GDPR has already established the framework, but no European body has yet dealt with a case of this magnitude. The question is not whether Brussels will respond, but when, with which instruments, and whether it will be quick enough to keep pace with companies like Meta, which are already rewriting the rules of the game.
