Anthropic updated its responsible scaling policy disclosures to state Anthropic cannot safely use Claude to fully autonomously design and build frontier models without meaningful human supervision