The Last Reversible Moment: A Constitution for Consequential AI

The Last Reversible Moment is a public declaration for consequential AI. It argues that before an AI-proposed action moves money, releases data, changes a record, deploys code, issues credentials, triggers a workflow, or otherwise creates a protected effect, the accountable receiving system must retain the final, current, and reviewable decision.

The declaration defines receiver-owned Action Acceptance at the last reversible boundary. One complete exact action is resolved against current authority, policy, evidence, declared effect count, remaining authorization capacity, and replay state, producing ACCEPT, HOLD, or REFUSE and a reviewable receipt. Capability, access, permission, and delegated authority may support the decision, but none is the final yes.

Every consequential action lands in someone’s life. In human terms, the last reversible moment is neighbor-love made operational: refusing to pass hidden uncertainty or avoidable harm to the person who bears the outcome. Proof before consequence is care made enforceable.

Version 1.0.1 is the current editorial reading edition. It adds the authoritative version DOI and a restrained mission statement centered on the person downstream. The declaration’s substantive principles, definitions, and architectural requirements remain unchanged from version 1.0.0.

Decision before action. Proof before consequence.