“Anthropic developed an AI model named Mythos that it deemed too dangerous for public release.”