It could be anarchy. It could be anything else.
The AI could simply do whatever its specialization is. It could be trying to direct the single-units or to disorganize them and try to keep some level of entropy. Maybe its own benefit would be a generalized state of chaos. Whatever.
The system itself could be considered the legal framework. The relationship between any supervisor and its units could be the metaphor for any subservient relationship in the real world, going from managed teams to parenting to politics to social dynamics between friends. It does not mean anything as long as the specifics of the system itself are not defined to permit the dynamics to be correctly mapped on some real-world dynamics.
Obeying the law? Are you obeying the law of gravity? Is it from your own will that you agree to stay on the ground?
So asking if that rule, that authority of the AI over the units is similar to human law, is I think a bit beside the point. It's more akin to the law of physics. It's simply the fabric of that reality.
Of course, you could then create a simulation that would try to recreate the same kind of domination as within a human society, but that would be only one kind of simulation, one kind of system, with specific parameters. It would not give much about the very essence of the system itself, only of it within constraints.
"Who is to say that the central planner must be the ontological equal of a single unit"
My impression is that's what most people here are discussing.
I can see a definition of an anarchy in your quorum example, can't you? But you say we could have AI which will find a balance. Interesting concept. Does it imply obeying AI <=> obeying the law?