CDT

What Kids and Parents Want: Policy Insights for Social Media Safety Features

Research with teens and parents identifies which safety features they actually want and use, arguing policy should center user-desired controls rather than blanket restrictions.

Content Moderation in the Global South: Four Low-Resource Languages

Synthesizes four case studies showing platforms overwhelmingly invest in English-language moderation, leaving low-resource-language speakers under-protected and subject to both over- and under-enforcement.

Moderating Quechua Content on Social Media

Quechua-language users face significantly worse experiences (mis-moderation, reduced reach, lack of language support) than Spanish-language posters.

Moderating Tamil Content on Social Media

Content policies are unevenly defined and enforced for Tamil speakers; automated moderation and Trust & Safety resourcing lag far behind English.

Moderating Kiswahili Content on Social Media

Kiswahili moderation suffers from limited language investment, sparse local context, and weak automated-system performance, harming users across East/Central Africa.

Real Time Threats: CSEA Prevention on Livestreaming Platforms

Finds real-time/ephemeral video creates distinct detection and enforcement challenges for child sexual exploitation prevention that existing moderation tooling is poorly equipped to handle.

Hated More: Online Violence Targeting Women of Color Candidates in the 2024 US Election

With University of Pittsburgh: women of color candidates experienced disproportionately high volumes of online abuse, hate, and mis/disinformation compared with other candidates.

Moderating Maghrebi Arabic Content on Social Media

Maghrebi Arabic is frequently mis-moderated, with both wrongful takedowns and failure to catch harmful content, driven by under-investment and ill-suited automated systems.

More Tools, More Control: Young Users on Handling Unwanted Messages

Young users want more granular controls to manage unwanted messages; current platform tools are insufficient, with design recommendations for youth safety that preserve autonomy.

Lost in Translation: Large Language Models in Non-English Content Analysis

Multilingual LLMs used for content moderation perform substantially worse in non-English (especially low-resource) languages; cautions against treating them as a fix for global moderation.