Anthropic Thinks Its Own Success Is Key to Making AI Safe