ANTHROPIC: "we may imminently meet our Responsible Scaling Policy’s definition of 'dramatically accelerated progress attributable to AI.'" "we do not anticipate being able to safely use Claude to fully autonomously design and build successive generations of frontier models"