Meta eases controversial employee-tracking programme, but here's the catch

Meta is scaling back parts of a controversial programme that records employees' computer activity to help train artificial intelligence models, but the company is stopping well short of abandoning the initiative altogether.

The project, known internally as the Model Capability Initiative (MCI), has become one of the most debated AI-related workplace policies in Silicon Valley. Designed to capture how employees interact with their computers, the programme collects data such as mouse movements, clicks and keystrokes in an effort to teach AI systems how skilled workers complete tasks.

According to a report by The Information, Meta has now introduced a handful of concessions following criticism from employees. However, the core premise remains unchanged. The company still intends to gather large volumes of workplace activity data from most of its workforce to improve future AI models.

Meta introduces limited safeguards to address employee concerns

Under the revised policy, employees will be allowed to temporarily suspend monitoring for up to 30 minutes if they need to attend to personal matters on their work devices.

More from Tech

The company has also created a limited exemption process. Certain employees will be permitted to request exclusion from the programme, including some remote workers facing internet bandwidth limitations, staff who regularly handle sensitive information, and those who work in environments where keeping devices constantly connected to power is difficult.

While these changes may ease concerns for a small group of workers, they do not fundamentally alter the programme's reach. Most employees are still expected to participate in the data collection effort.

Meta has also reportedly addressed one of the more practical complaints surrounding the initiative. According to Reuters, the company told workers that it has improved the software's battery efficiency after concerns emerged that the monitoring tools were placing additional strain on laptops.

The programme reflects Meta's broader push to accelerate AI development at a time when competition among major technology companies is intensifying. Company leadership believes observing how experienced employees perform digital tasks can provide valuable training material for future AI systems.

Internal backlash grew as Meta doubled down on its AI ambitions

Resistance to the initiative has been building since its introduction last month.

The rollout came shortly before Meta announced significant organisational changes, including the elimination of around 8,000 jobs and the reassignment of thousands of employees into AI-focused roles. For some workers, the timing intensified concerns about workplace surveillance and the growing emphasis on artificial intelligence across the company.

Meta CEO Mark Zuckerberg has publicly defended the programme in conversations with employees. In leaked audio from a company-wide meeting, he argued that observing highly skilled workers represents one of the fastest ways to improve AI capabilities.

"Watching really smart people do things" is the best way for AI models to improve quickly, Zuckerberg said, adding that the company's workforce possesses a level of expertise that would be difficult to replicate through conventional data collection methods.

Addressing concerns that the initiative could be used for performance evaluation or surveillance, Zuckerberg insisted that the purpose was purely related to AI development.

"None of the data is being used for, like, looking at what people are doing, or surveillance, or performance track[ing], or anything like that," he said. Instead, he argued, the information would help AI systems learn how experienced professionals use computers to complete complex tasks.

Zuckerberg also suggested that if the initiative proves successful, Meta could expand similar efforts in the future.

The latest revisions may reduce some employee frustration, but they are unlikely to end the debate. As technology companies race to build more capable AI systems, questions about how training data is collected, and how much monitoring workers should accept in the process, are becoming increasingly difficult to ignore.