The Terminus Doctrine

Article I

Disqualification

The machine is not equipped to be the terminus of a consequential decision. Deciding requires faculties it does not have:

• It cannot verify in real time — it acts on a frozen model with no live channel to the world.
• It cannot read tone — it holds the words, not the register; the plea and the routine request are the same token-stream at the same confidence.
• It resolves to a probability, not a judgment — an output that sounds like a decision and is a distribution wearing a decision's clothes.

These are not limitations to be engineered away. They are a structural disqualification — so the decision was never legitimately the machine's.

Article II

Compute, not Decide

The line is the terminus, not involvement. The machine may triage, sort, surface, and recommend — that is the real and valuable work. It may not be the place a consequential decision closes.

Recommend: yes. Decide: no. A system that merely assists is permitted; a system that is the terminus of a consequential call is not — regardless of what it is named.

Article III

The Hand-Off

The machine's duty is to deliver the consequential decision to a human — intact, in time, and attributably. There, the accountability systems that already govern human decisions take over: a named human decides, owns it, and is held to it by the same machinery that governs every human who decides badly.

This is not a new court. It is the delivery that guarantees the decision reaches the one place the existing courts can grab it.

Foundation

Not Novel

This is the refusal to exempt automated processes from the accountability that already governs bounded, human-authored programs — extended at two seams only: nondeterminism (answered by determinism-or-a-human-apex) and scale (answered by audit and appeal). The claim that AI needs wholly new governance is itself the dodge; it manufactures the gap the evasion lives in.