Meta’s latest AI improves its terrible content moderation

Meta has announced a global rollout of AI-powered content moderation tools across its apps, reporting early results including detection of 5,000 daily phishing attempts, an 80% reduction in fake celebrity profile reports, and doubled detection of adult content violations. The company claims AI outperforms its human moderation teams on these tasks. The Register notes some of these capabilities — like detecting impossible logins from different locations — have existed in enterprise security tools for years, suggesting Meta's human moderation was simply not using available techniques. The announcement coincides with Meta shutting down its Horizon Worlds metaverse platform after spending over $80 billion on the initiative, while pivoting toward developing 'superintelligence.'