“AI models from Anthropic and other major labs are currently tested by third-party organizations, including government bodies, for dangerous capabilities related to national security, such as biological weapons development and cyber offense.”