Meta has begun installing surveillance software on every US employee's work computer that records mouse movements, keystrokes, clicks, and periodic screenshots. The program, called the Model Capability Initiative, feeds this data directly into training AI agents designed to replicate how humans interact with computers.

Employees cannot opt out. When staff asked whether they could decline, they were told: "No, there is no opt-out on your work-provided laptop." The only escape route, as Platformer reported, is to relocate to Europe, where EU privacy laws prohibit this kind of workplace surveillance.

This is happening alongside Meta's announcement that it will lay off 10 percent of its global workforce later this month. The employees being surveilled are, in many cases, training the systems that will eliminate their own positions.

The Data Gap That Started This

The initiative traces back to a specific problem. Agentic AI models struggle to replicate how humans actually use computers. They can write text and generate code, but they cannot reliably navigate interfaces, switch between applications, or perform the kind of multi-step computer work that knowledge workers do all day.

Meta paid $14 billion last year for a 49 percent stake in Scale AI, installing co-founder Alexandr Wang as head of Meta's superintelligence team. Wang's core problem has always been data. As he said in 2024: "There's no pool of really valuable agent data that's just sitting around anywhere." Meta found a pool: its own employees.

With roughly 80,000 employees working daily on company-owned computers, Meta has access to an enormous, constantly refreshing dataset of exactly the kind of human-computer interaction that agentic models need. Internal memos seen by Reuters are explicit: "This is where all Meta employees can help our models get better simply by doing their daily work."

Agents to Find Agents to Rate Agents

The New York Times reported Thursday that Meta's internal AI push has reached a point where employees are being told to build so many AI agents that "others had to introduce agents to find agents, and agents to rate agents." The result, according to current and former employees, is widespread anger and anxiety.

Some employees told the Times they no longer see Meta as a place for a long career. Others are actively looking for new jobs. Some are trying to signal that they want to be laid off so they can collect severance pay rather than continue building their own replacements.

Meta has rebranded its enterprise AI product from "AI for Work" to the "Agent Transformation Accelerator." The surveillance data being collected from employees will train agents that Meta plans to sell to other employers. The company is not just replacing its own workforce. It is building the product to replace everyone else's.

The EU Exemption Tells You Everything

European employees are exempt from the surveillance program. The GDPR limits when personal data can be used to train AI systems, and the EU AI Act bans practices like inferring employee emotions from workplace monitoring. Meta knows this monitoring is not permissible under European law, so it simply does not do it there.

The US has no equivalent protection. American employees have essentially no legal recourse against employer surveillance of company-owned devices. Meta has framed the data collection as part of employees' existing job duties, which means no additional consent is required.

Law professor Ifeoma Ajunwa has described this dynamic as "captured capital": the involuntary collection of worker data by companies to automate workplaces, leading to worker displacement and reinforced employer control. The workers producing the most valuable training data are the ones whose jobs will be automated first.

The Irony Is Not Lost

Meta spent twenty years harvesting the personal data of billions of platform users without meaningful compensation. That its own highly paid employees are now subject to a version of the same extraction has, as TechPolicy.Press noted, elicited schadenfreude in some quarters. But the implications reach far beyond Meta's campus. If this model works, every company with employees on company devices has the same option.

Reporting from the New York Times, Reuters, The Verge, Platformer, and TechPolicy.Press contributed to this article.