The Bluesky post that reframed AI governance as a constitutional problem [3] is doing something specific: it removes technical alignment from the center of the argument. The claim is not that labs are building unsafe systems, but that no institution with democratic legitimacy currently holds the authority to compel a different choice. That is a structural claim about power, and it changes what a solution looks like. Safety research answers the question of what a well-behaved system does. It does not answer who gets to define "well-behaved" and enforce that…