Chinese censorship is leaking into answers from American AI

'Multi-part case study on China’s media' finds that AI models can't hallucinate away Chinese censorship |…

Nature: 3M state-media docs in AI training data; Claude, GPT-4o reproduce Beijing phrases in 3–10% on China. Meta audit shows US models refuse political criticism 34% more often in authoritarian countries, signaling authoritarian information controls reach globally-deployed AI systems.