Xi Jinping's minority crackdown has nothing to do with AI, and everything to do with what AI companies ignore at their peril
China's new 'ethnic unity' law forces 125 million minority citizens to abandon their languages and cultures. For AI firms doing business there, the political climate just got harder to ignore.
Key points
- China's new ethnic unity law compels roughly 125 million people from 55 minority groups to adopt Mandarin and conform to Han Chinese cultural norms.
- Xi Jinping frames the policy as national integration; Tibetans, Uyghurs and others call it forced assimilation.
- Minorities represent about 9% of China's total population.
- The policy tightens a pattern of centralised control that directly shapes what AI tools can say or show inside China.
- Western AI companies operating in or selling to China face growing pressure to build products compatible with these restrictions.
China's government has passed a sweeping new law requiring the country's ethnic and religious minorities to unify around a single national identity, defined by the Communist Party and expressed almost entirely in Mandarin. The Guardian reported the details of what Xi Jinping calls his sinicisation policy, a word meaning "to make Chinese" in the mould of the dominant Han ethnic group.
Roughly 125 million people belong to 55 recognised minority groups, making up about 9% of China's population. Under the new law, compulsory Mandarin instruction expands, and local languages taught in schools for generations face restriction or removal from classrooms.
Xi has described the goal as minorities "hugging tightly like pomegranate seeds" to build a unified China. Critics, including human-rights groups, use a blunter phrase: ethnic cleansing by bureaucratic means.
Why does this matter for AI?
Every AI product that operates inside China must comply with Chinese law, and Chinese law is tightening fast.
Large language models, the technology behind chatbots like ChatGPT and Claude, are trained on enormous amounts of text. That text shapes what the model knows, what it'll discuss, and what it refuses to say. Inside China, regulators already require AI systems to reflect "core socialist values" and avoid content the Party deems divisive. We covered China's broader AI governance ambitions on 20 July 2026, when Xi used the World AI Conference in Shanghai to pitch cooperation while warning the West against security overreach.
A law restricting minority languages doesn't stay inside school buildings. It shapes the data available for training Chinese AI systems, and determines which voices those systems reflect and which they erase.
Western AI companies want access to China's market of more than a billion people. That access has a price: products shaped to fit a political environment that's grown measurably more restrictive.
What does this mean for ordinary users outside China?
If you live outside China, no AI product you use today is directly bound by this law. But the global AI industry isn't neatly divided by borders.
Models trained partly on Chinese data, or built by companies with Chinese investors, carry the assumptions baked into that data. A Tibetan dialect, a Uyghur folk tradition: if these disappear from Chinese digital life, they become harder for any AI to understand or represent accurately.
| Group | Estimated population | Primary concern |
|---|---|---|
| Uyghurs | ~12 million | Language, religious practice |
| Tibetans | ~7 million | Language, cultural identity |
| Mongolians | ~6 million | Language instruction in schools |
| Other minorities | ~100 million | Varying degrees of assimilation pressure |
The honest read here: when you use an AI tool, the politics of where its training data came from are already inside the product. You can't see them. That's worth knowing.
