Anthropic said on Friday that the U.S. government has allowed it to release its powerful Claude Mythos 5 artificial intelligence model to some "trusted" U.S. organizations, partially reversing an order two weeks ago to suspend access over national security risks.