The Workers Training Meta's AI Are Being Replaced by It

Over 700 Covalen workers in Dublin who refine and moderate Meta’s AI models are facing layoffs — a direct consequence of Meta’s accelerating bet on AI automation replacing human content enforcement. The cuts mark the second significant workforce reduction at the Dublin contractor in under six months.

Meta’s AI Automation Displaces Its Own Trainers

Around 500 of those at risk are data annotators who teach Meta’s AI models to recognize harmful content by flagging violations and modeling correct decisions. The irony is not lost on those affected. “It’s essentially training the AI to take over our jobs,” one anonymous Covalen employee told Wired. “We take actions as the perfect decision for the AI to emulate.”

According to Wired, workers were notified via a brief video call on Monday with no opportunity to ask questions. Nick Bennett, one of those on the call, said: “We had a pretty bad feeling [before the meeting]. This has happened before.” Bennett described the day-to-day work as psychologically taxing — “quite a grueling job” that involves simulating bad actors to probe AI guardrails against the most disturbing content categories.

The Economics of Vendor Displacement

Meta’s reasoning is explicit. Spokesperson Erica Sackin confirmed the company is “deploying more advanced AI systems to transform our approach to content enforcement,” adding that Meta will be “reducing our reliance on third-party vendors.” Workers were told cuts stemmed from “reduced demand and operational requirements” — language that obscures a strategic pivot.

This is the second wave of Covalen cuts in recent months. A November 2025 round of approximately 400 job losses sparked a worker strike, foreshadowing the broader instability now reaching hundreds more.

The macro picture: Meta announced plans to cut one in ten employees company-wide while nearly doubling AI infrastructure spending. In January, CEO Mark Zuckerberg declared that “2026 is going to be the year that AI starts to dramatically change the way that we work.”

Why This Matters

The Covalen situation exposes a core paradox of generative AI: the technology depends on human labor to become safe enough to replace that same labor. Data annotation — low-paid, psychologically demanding, largely invisible — has been foundational to every major AI model. As automation matures, the workers who enabled it are among the first displaced. This dynamic will increasingly challenge policymakers and labor advocates as AI capability and corporate consolidation accelerate.