“Anthropic believes that developing "scalable supervision" for AI is the key lever that will enable the confident and continued improvement of AI capabilities.”