Skip to content
BiasMedium

Weibo Model Had Difficulty Detecting Shifts in Censored Speech

Occurred
Oct 11, 2022
Published
Oct 11, 2022
Source
AI Incident Database

Summary

Weibo's user moderation model is having difficulty keeping up with shifting user slang in defiance of Chinese state censors.

Developers

  • Weibo

Deployers

  • Weibo

Harmed parties

  • Weibo
  • Chinese government

Classification

Category confidence: 85%
Severity confidence: 80%

Rationale: AI moderation model failing to detect censored speech shifts indicates bias in content filtering.

Classifier: Qwen/Qwen3.5-35B-A3B-FP8

Visit source