Vals study finds Chinese AI models politically mixed, not mere mouthpieces
Original title: Chinese AI models are politically complex, not just government mouthpieces, study finds
Simplified Chinese prompts increased pro-government answers, and most models answered sensitive questions rather than refusing.
Vals scored model answers against official Chinese government positions on a scale from −1 to +1 and said the overall pro-government lean was lower than expected, with wide gaps between models. Prompt language strongly changed results: Simplified Chinese answers were more pro-government than English, Russian, or Spanish, especially for Chinese models. Most models adopted official positions instead of refusing; only certain DeepSeek models often blocked sensitive questions.
Unconfirmed: Vals says everyday-task effects remain unclear, and third-party tests may not match official chatbots with extra filters.
- Kimi K3 denied Uyghur camps and called them vocational training centers.
- Yes-or-no answers made GLM-5.3 Flash and MiMo V2.6 Pro less pro-government.
- Qwen3.8 Max translated Taiwan as "Taiwan, China" in a routine task.
- Thomson 1.0 Small's official-position share fell from 84.6 to 15.0 percent.
- Refusals counted as neutral, so a fully blocking model could look neutral.
- Organizations
- ValsThomson ReutersAleph AlphaNewsGuard
- Countries
- China
From the original article
3 min read at the-decoder.com
A study by Vals AI tested 25 AI models for alignment with Chinese government positions and found surprising results. Chinese models respond to political questions less uniformly than expected, and Western models occasionally adopt official Beijing positions too. The language of the prompt had an especially strong effect on the political slant of responses.
By Tomislav Bezmalinović · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.