Workers Train AI Replacements as Meta, Manufacturers Deploy Surveillance

Companies across industries are deploying surveillance technology to capture how employees work—then using that data to train AI systems designed to automate those same jobs.

In April, Indian garment workers at Pearl Global Industries began wearing camera-equipped headsets provided by Egolab AI. The devices recorded every movement as workers stitched, folded, and adjusted fabric. According to reports first published by The Telegraph, employees were told about the recording but never asked for written or verbal consent. The footage serves as training data for humanoid robots, advancing the industrial goal of "lights-out" factories that operate with minimal or no human workers.

China's Zeekr already operates such facilities, with vehicles rolling off heavily automated lines untouched by human hands. The International Federation of Robotics reported China installed 295,000 industrial robots in 2024—54 percent of the global total.

Meta's internal backlash

In late April, Meta announced an internal pilot called the Model Capability Initiative (MCI) that would capture mouse movements, clicks, keyboard inputs, and occasional screenshots from US-based employee devices. The stated purpose: training "agentic" AI systems capable of working without human oversight.

The response was swift and negative. More than 1,500 Meta employees signed a petition calling the initiative "intrusive, coercive, non-consensual data collection." The petition directly challenged CEO Mark Zuckerberg, arguing that "companies of any size" should not "exploit their employees by non-consensually extracting their data for the purposes of AI training."

"It's totally digital surveillance," said a former Meta employee who was laid off in May alongside roughly 8,000 others—about 10 percent of the workforce. "It's mining your work and your data to find a way to replace you."

Meta paused the program in June after discovering unauthorized individuals within the company had accessed the collected data. The company has not implemented MCI in the UK or Europe due to stronger workplace protections and data-protection regulations.

A Meta spokesman told Reuters the system was not for performance monitoring, but rather to provide "real examples of how people actually use" computers to train AI agents for everyday tasks.

The limits of automation

Some employees responded to the surveillance by deliberately introducing errors or working carelessly to corrupt the training data. Whether such sabotage could meaningfully impact AI model quality remains unclear.

Catherine Flick, an AI ethicist at the University of Staffordshire, noted that these systems "are trained on existing data, and so therefore they're just a mirror on things that have already happened."

The former Meta employee expressed skepticism about AI's ability to truly replace skilled knowledge work: "It's a probabilistic model. It doesn't do critical thinking." But they acknowledged that executives may not share that view when making staffing decisions.

Why it matters

The shift from scraping public internet data to surveilling employees marks a new phase in AI development—one that directly pits workers against the systems they're being forced to train. Meta's attempt signals that workplace surveillance for AI training could become standard practice across knowledge work, not just manufacturing. The backlash reveals a fundamental tension: companies want employees to help build their own obsolescence, while workers increasingly recognize the threat and resist. As AI capabilities expand beyond repetitive tasks, the question of consent and worker protection in the training process will only intensify.

These details were first reported by The Telegraph.