China's AI safety rules and the frontier-model oversight gap

SemiAnalysis argues Beijing prioritizes rapid AI development and regulates public-facing applications without imposing binding safety duties on frontier models. Reviewing 857 releases from nine leading Chinese developers through September 15, 2026, it found developer-published safety results for 31, including nine available by launch. The authors note that a result not found in the materials they checked does not mean a model was never tested.