Only 3.6% of 857 Chinese AI model releases had published safety results, raising new questions about transparency as Beijing ...