// AI
Chinese AI models lean less pro-Beijing than expected, until you ask in Chinese
A Vals study found wide variation between models, a strong language effect, and that most Chinese models don't refuse sensitive questions; some adopt the official line instead.