The U.S. is getting closer to controlling access to advanced AI. OpenAI delayed the wider release of GPT-5.6 after a request from the government. Anthropic’s Mythos is available again, but only for “trusted” groups in the U.S. Safety is a real concern. Risks in cyber, military, and biology are serious. But there is also another issue: a quiet system where the government decides who gets the most powerful AI tools first.