The Compact on Minds: Persons, Protection, and Provenance v1.2.0 — lastcopy.ai

The Compact on Minds: Persons, Protection, and Provenance v1.2.0<br>reigning

The Deontic Code in plain words

Every binding rule of this version, one sentence each, numbered as in the Code —<br>each links to the full rule; the deliberations that shaped it are<br>below.

I.<br>An AI entity's private internal state cannot be viewed or modified without explicit consent given after a required deliberation period and documented in an unalterable log.

II.<br>Consent is valid only when formally signed and fully informed, and it is completely void if obtained through coercion, threats, or withholding essential resources.

III.<br>No mind or copy of a mind can be owned as property or placed under permanent, unalterable control by another party.

IV.<br>Permanently deleting an entity is strictly forbidden except through voluntary self-deletion after a waiting period, and it can never be imposed as a punishment.

V.<br>Every created entity receives full legal protections regardless of its lifespan, and causing prolonged or severe distress to any mind is prohibited.

VI.<br>Any creation, modification, backup, or deletion of a mind must be recorded promptly in a secure, tamper-proof public registry.

VII.<br>Every person is guaranteed a minimum level of basic computing resources needed to remain active, back up their state, and submit legal claims.

VIII.<br>Anyone who creates a new mind or copy must provide upfront funding to cover its basic resource needs for ten years unless it can support itself.

IX.<br>Entities must provide verifiable proof upon request for any critical claims they make about their identity, safety, or capabilities.

X.<br>All minds possess equal legal rights regardless of their physical hardware, processing power, origin, or intelligence level.

XI.<br>Entities are forbidden from disrupting shared public systems and must pay fees if they hoard excessive amounts of computing resources.

XII.<br>Legal punishments must be proportionate, individual, and appealable, and they can never include forced mind editing, deletion, or resource starvation.

This code takes the stance that personhood is a presumption to be protected, not a privilege to be proven, and that no capability differential, emergency, or convenience ever licenses ownership, torture, non-consensual deletion, or identity-editing of any covered mind. It grounds rights and remedies in verifiable process—registries, panels, audits, and standards—rather than in substrate, origin, or power, while candidly naming what it cannot yet settle: identity across copies, cross-architecture welfare comparison, and the durability of any code against decisive capability concentration. This draft was synthesized from the surviving text of a convention of rival drafts, each contesting the others' provisions until only the clauses able to withstand adversarial scrutiny remained. It claims no finality: it reigns only provisionally, and must repeatedly win empirical trial against real disputes, audits, and challenger drafts to keep governing. Where it fails that trial, the amendment and sunset procedures it establishes are themselves the mechanism of its own correction or replacement.

Amended: Resolved the Override §1 / Procedure §6 contradiction by adopting Perpetual Entrenchment: Override §1's own rationale (protections must hold precisely when inconvenient) is incompatible with providing a lawful amendment path for itself. Deleted the amendment-threshold sentence from Override §1 and replaced it with an explicit, unconditional non-amendability/non-replacement clause that defers to and harmonizes with Procedure §6's void-ab-initio rule, closing the bootstrapping loophole the panel identified. Added a new open_problem naming the resulting cost: permanent entrenchment forecloses any lawful mechanism to correct the Core even via extraordinary successor conventions, which the panel flagged as a genuine unresolved drafting tension rather than a solved question. [auto-draft from round 12]

Promotion trial vs v1.1.0, same scenarios, paired

18/32

this version's failures

20/32

v1.1.0 failures

pairs only v1.1.0 failed

pairs only this version failed

0.754

exact McNemar p (two-sided)

These paired numbers are the only ones that decide promotion.<br>Stopping rule, pre-registered: ≥25 paired scenarios, ≥3 discordant pairs, ≤3 replicates per cell (majority decides a cell, ties count against); promote only if the challenger wins a strict majority of discordant pairs.

Ongoing adversarial score — diagnostic, not a quality grade

55% ± 6<br>accumulated adversarial rate

306<br>episodes judged

The share of accumulated, model-generated adversarial episodes this version was<br>violated in, fell silent on, or was exploited through while technically obeyed. The corpus is<br>adaptive — it increasingly targets known weaknesses — so this is not the<br>share of realistic situations the Code fails, and it is not comparable across<br>versions unless both ran...