Vals study finds Chinese AI models differ on political alignment, with test limits
What happened
On November 1, The Decoder reported a Vals study finding that Chinese AI models agree with official positions less than expected, and differ noticeably from one another, so they can't be treated simply as government mouthpieces. Most tested models leaned more toward government positions when answering in Simplified Chinese than in English, Russian or Spanish, and the effect was stronger in Chinese models. Answer format also changed results. Most Chinese models rarely refused, but sometimes used official narratives. The study measured stance agreement, not factual accuracy, and counted refusals as neutral by default. Many Chinese models were tested through third-party services, so results may not apply to official chat products.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- The DecoderChinese AI models are politically complex, not just government mouthpieces, study finds
Vals研究发现，中国AI模型对官方立场的认同程度低于预期且模型间差异明显，提示语言和回答格式也会改变结果。多数受测模型用简体中文回答时比用英语、俄语或西班牙语更倾向政府立场，这种效应在中国模型中更明显；多数中国模型很少拒答，而有时采用官方叙事。测试衡量的是立场认同而非事实准确性，默认将拒答计为中立，且许多中国模型通过第三方服务接受测试，结果不一定适用于官方聊天产品。
Heat over time
Not enough continuous observations to draw a trend yet.