Dario Amodei published an essay saying the government should have the power to block AI models with unacceptable risks. Within 48 hours, the government did exactly that — to Anthropic's own models. The lesson: if you invite government into the room, it may not leave.