In its S-1 initial public offering filing, Anthropic stated that the company does not believe it can safely conduct recursive self-improvement on a fully automated basis