Meta's three-month-old Applied AI unit faces internal crisis as engineers protest forced reassignments
Thousands of Meta engineers drafted into a data-labeling unit report morale collapse, with some calling the work 'soul-crushing' and describing their assignment as involuntary.
Last verified:
Forced data-labeling assignment triggers employee revolt at Meta
Meta has conscripted approximately 6,500 engineers and product managers into a newly created Applied AI division tasked with generating training materials for the company’s AI research, according to reporting by TechCrunch AI and Wired. The wave of involuntary reassignments, which began in April, has triggered significant morale collapse within the unit, with employees describing their circumstances using terms like “draftees” and calling the work environment oppressive.
According to TechCrunch AI, the discontent boiled over this week when an unidentified person hijacked a livestreamed employee-only presentation and delivered an expletive-laden tirade directed at a senior Meta AI executive, prompting at least one presenter to cover their face during the outburst. The incident crystallizes weeks of simmering dissatisfaction within the three-month-old Applied AI team.
Assignment structure and justification
Meta’s reasoning for reassigning internal staff rather than hiring external contractors centers on perceived cognitive ability. According to a leaked audio recording from an April internal meeting reported by TechCrunch AI, CEO Mark Zuckerberg explained that the company believes its average employee possesses substantially higher intellectual capacity than third-party vendors. Alexandr Wang, who joined Meta as chief AI officer after selling his data-labeling firm Scale AI to the company for $14.3 billion, apparently advocated for leveraging internal talent for this labor-intensive task.
According to Business Insider, Meta’s internal April announcement indicated the company’s AI models were underperforming humans on technical problem-solving. The company determined it needed real examples of how humans actually use computers to complete routine work—the data the Applied AI unit now generates.
Escalating workplace friction
The assignment process itself triggered immediate resistance. According to TechCrunch AI, many employees learned they were being moved into the unit through surprise emails with little choice to decline. On Reddit, self-described reassigned engineers described the notification process as essentially random.
Beyond the Applied AI unit, broader morale challenges plague Meta. TechCrunch AI reports that more than 1,600 Meta employees company-wide have signed a petition against a keystroke and click-monitoring program designed to capture AI training data. Meta Chief Product Officer Chris Cox recently addressed employee concerns on a call, acknowledging the workplace environment has grown increasingly harsh.
Why This Matters
Meta’s internal labor mobilization strategy reveals a critical tension in scaling AI training infrastructure: the company lacks sufficient external data-labeling capacity or trusts it inadequately to hit its model development targets. By conscripting 6,500 salaried engineers—whose primary expertise lies in systems design, not data annotation—Meta is trading short-term training velocity for long-term retention risk. If Applied AI team attrition accelerates or the quality of engineer-generated training data underperforms the company’s projections, Meta may face delays in Superintelligence Labs’ agent development roadmap. The broader petition against monitoring suggests Meta’s workforce is drawing a line on data practices, a signal that future reassignments or surveillance expansion could encounter organized resistance.
Frequently Asked Questions
Why did Meta reassign thousands of engineers to this new unit?
According to Meta CEO Mark Zuckerberg in a leaked April audio, the company believes its own employees have significantly higher cognitive ability than external contractors, making them better suited for generating coding problems and real-world task examples to train AI agents.
How did employees find out they were being moved?
Many learned through surprise internal emails in April, with some describing the process as random and involuntary—employees felt pressured to accept reassignment or leave the company.
What specific work are these engineers doing?
They are creating puzzles, coding challenges, and documentation of real user workflows to serve as training data for Meta's AI models, particularly to improve performance on technical tasks.
Is there broader employee discontent at Meta?
Yes; over 1,600 Meta employees have signed a petition against a monitoring program tracking user clicks and keystrokes for AI training, and leadership has publicly acknowledged a deteriorating workplace atmosphere.