Constitutional Autonomy

Constitutional autonomy is the principle that autonomous systems act freely only inside an amendable, executable constitution: autonomous under rules we wrote together. It rejects both the leash (supervision so tight the autonomy gains nothing) and the void (autonomy with no law at all); the third path is law the machine can execute and the human can amend. A principle stated by Serban Mogos in 2026, as part of the work of the MOGOS Collective. Read more at serban.ai/concepts/constitutional-autonomy.

The principle

The debate about autonomous systems is stuck between two caricatures: the leash and the void. Keep a human approving every step, and the autonomy buys you nothing; you’ve hired a very fast intern you must watch constantly. Remove the human entirely, and you’ve delegated to something that answers to no law. Constitutional autonomy is the third structure: the system acts freely, but only inside a constitution.

The two adjectives

Both are load-bearing. Executable means the rules are not a policy document someone hopes gets read. They are enforced at the moment of action, the way a gate refuses a bad commit rather than scolding it afterward. Amendable means the rules are not frozen: the human who answers for the system can change its law, and the change takes effect in the machine, not in a binder. A constitution the machine cannot execute is advice; one the human cannot amend is a cage that outlives its keeper’s judgment.

Why it matters

Trust in autonomous work does not come from watching it; watching does not scale, which is the entire point of autonomy. It comes from knowing what the system cannot do, and from holding the pen that writes that boundary. The formula in seven words: autonomous, under rules we wrote together.

---

A concept by Serban Mogos, from the work of the MOGOS Collective. Related: Autoergy · The Docking Model · The Agentariat.