Meta’s AI moderators cut error rate 13% — but staff fear chaos
meta replaced 50% of human moderation with llms and targets 90% by year-end, causing internal warnings about rushed oversight and layoffs despite improved accuracy stats.
meta shifted 50 percent of its moderation workload to llms this year and targets over 90 percent automation for some content types by the end of 2025. the company reports its new muse spark model, which replaced google’s gemini for these tasks, makes 13 percent fewer errors and detects 10 percent more violations than human moderators. however, insiders argue the rollout is dangerously rapid, citing instances where ai incorrectly penalizes harmless content due to insufficient oversight. this transition directly impacts external contractors through layoffs while meta bets billions on autonomous systems to handle nuanced policy enforcement across languages.