Policy · 1h ago
Automated Moderation Fails Marginalized Languages, EFF Report Finds
EFF's report shows Meta's automated moderation deletes nonviolent Arabic content 77% of the time and misses hate speech. Five years later, overzealous moderation persists, especially for low-resource languages like Maghrebi Arabic and Kiswahili. The report calls for transparency, cultural competence, and stronger appeals processes.
Meridian48 take
The piece rightly highlights systemic bias in automated moderation, but understates the business incentive for platforms to prioritize scale over accuracy.
content-moderationai-bias