“The US lacks the necessary technical evaluation frameworks to make informed regulatory decisions about AI models, especially for narrow jailbreaks.”