Meta pushes AI moderation to 90% by year-end

What happened

Meta has already replaced roughly half of all human moderation requests with large language models in 2025 and plans to push that share above 90 percent for some content types by the end of the year. The company switched from using Google's Gemini for moderation to its own new foundation model called Muse Spark. According to tests since March, Meta's language models make 13 percent fewer errors than humans when enforcing content policies while catching 10 percent more actual violations.

Why it matters

The shift is expected to save the company billions annually. However, Meta employees say the models still remove or shadow-ban harmless content and there isn't enough oversight for such a rapid rollout. The transition is already leading to layoffs, especially among external contractors. Unlike traditional ML classifiers that struggle with satire or evolving language, the language models are supposed to better grasp nuance and cover more languages—but the real-world tradeoff between speed and accuracy is a live concern for how content moderation will work.

What to watch

Meta disputes the cost argument and points to quality instead, emphasizing that its tests show improved performance. The core tension is whether the company's internal quality metrics reflect what actually happens when the AI moderates at scale across diverse languages and content types.