The White House and Anthropic are working on a framework that would assess the severity of security flaws in new AI models and guide potential government intervention, Politico reports