“It will be impossible to definitively prove that Chinese-developed AI models are free from hidden vulnerabilities or malicious tool-use capabilities.”