According to CoinMeta, Anthropic indicates that it may soon meet the definition of "significantly accelerated AI progress" in its security policy. This disclosure is made in its revised IPO document, based on internal estimates and expectations for future models, rather than the current model. Anthropic also states that no current model has yet exceeded this threshold, and they are evaluating what it means to meet the AI research and development capability threshold under their responsible expansion policy. In addition, Anthropic expects that it will not be safe to allow Claude to completely autonomously design and build subsequent cutting-edge models without meaningful human supervision, and they are considering policy updates to maintain human control. According to the S-1 /A document, Claude contributed over 90% of the code to the Anthropic repository in September.