Platforms including YouTube, TikTok, and X are using large language models to scan content for known falsehoods, flag suspicious patterns, and route borderline posts to human reviewers. The systems can process millions of posts per hour, a pace no human moderation team could match. Early results show meaningful reductions in the spread of specific categories of health misinformation.
However, AI moderation systems carry their own risks. Multiple studies have documented cases of over-moderation where satire, minority languages, and legitimate political dissent were incorrectly flagged and removed. Civil liberties groups argue that delegating speech decisions to AI systems without sufficient human oversight creates new forms of censorship that are harder to challenge and appeal.
