“A dangerous AI model, such as one developed by a company like Mistral AI, could be designed in one country and then used by an actor in a second country to attack a third country.”