Anthropic signaled that broader access to its controversial Claude Mythos AI models may arrive "in the coming weeks" after limited testing.