US-developed AI models also have inherent risks when considered in isolation