The Dire Defect of ‘Multilingual’ AI Content Moderation

Social media companies claim new language models can remove harmful content in every language. But those systems’ shortcomings can have vast consequences.