Meta's AI Moderation Catches 97% of Hate Speech in Real Time
October 30, 2025 | The Verge
Meta has upgraded its AI moderation to detect 97% of hate speech across 50+ languages in real time—up from 83% in 2024.
The system uses multimodal analysis of text, images, and video context.
Key improvements:
- Detects dog whistles and coded language
- Understands regional slang and memes
- Reduces false positives by 40%
Still, human reviewers handle 3% of edge cases.
“Scale requires AI, but trust requires humans,” said Meta’s AI ethics lead.