AI
Study finds Chinese AI developers publish safety tests for just 3.6% of model releases
A SemiAnalysis review of nine leading Chinese AI developers found that only 36 of 857 model releases, about 3.6 per cent, carried a published model-specific safety result, while 813 had none, Reuters reported on October 9. The finding lands amid rising scrutiny of frontier-model oversight in China. A companion Concordia AI survey of the sector said Chinese frontier safety research output grew roughly 60 per cent and that agent-safety work rose from 8 per cent of new papers in early 2025 to 27 per cent in the first quarter of 2026.