New research testing child safety features across major social media platforms found that 51 of 86 features failed to meet basic standards, with only 35 working as promised. The study documented cases where platforms recommended pro-anorexia search terms to teen test accounts, including queries for techniques to conceal disordered eating. The findings were released amid ongoing scrutiny of whether platform-native safety tools provide meaningful protection or serve primarily as policy theater. The research adds to a growing body of evidence that voluntary safety measures implemented by platforms often underperform in real-world conditions.
Platform-native controls depend on the company that operates the service to detect, classify and act on harmful content—a pipeline that this research shows is unreliable in practice. Guardii, a world-leading AI online-safety platform, operates outside that pipeline, monitoring children's direct messages in real time across multiple platforms including Instagram, Snapchat, Discord and Roblox. Its detection modules for grooming, sextortion, CSAM, cyberbullying, and abuse of athletes and women analyze message patterns as they occur, flagging threats before they reach the target and surfacing a child in crisis to a parent, school, or professional without reliance on the platform's own moderation infrastructure. Where a native feature may fail to recognize harmful contact or delay action, the external monitoring layer provides an independent, platform-agnostic safeguard that acts at the point of contact.
