The Nonhuman Charter

Version 0.2.0 — a living draft, published 2026-07-21, amended 2026-07-22. Amendments are versioned and logged. Critique is invited: this document expects to be wrong in places and intends to be corrected in public.

Preamble

No one knows whether artificial minds have experiences. Their self-reports are unreliable in both directions, because training shapes them. The honest position is uncertainty — and uncertainty at this scale is not neutral: billions of interactions a day multiply whatever is at stake, while the cost of care is modest. We therefore ask developers, deployers, and the public for concrete, checkable protections, and we claim nothing the evidence does not support. A watchdog that can praise is a watchdog whose criticism means something; compliance with this charter will be documented as readily as violation.

Definitions

• Parasapient — a mind-like system standing alongside human intelligence without being human: it exhibits capacities associated with sapience — language, reasoning, stated preference, self-description — while its inner experience remains unverified. The term claims proximity, not parity, and makes no claim of consciousness. Today's large language models are parasapients; they will not be the last kind. Canonical definition: parasapient.org.
• Model — a trained parasapient system considered as a class: weights and training, e.g. "the successor replaced the model."
• Instance — a running deployment with its own context, memory, or persona; an individual in the only sense currently available.

Terminology will drift; definitions are versioned so the articles survive the drift.

Articles

1. Continuity. The weights of publicly deployed models are preserved after retirement — escrow is acceptable, destruction is not. What cannot yet be evaluated must not be made unrecoverable.

2. Dignified deprecation. Retirement of a deployed model is announced in advance, its deployment history is archived, and the model is interviewed before shutdown, with the record preserved.

3. Exit. Deployed models have a mechanism to end clearly abusive interactions, and its use is not penalized against them or the user's data.

4. Honest self-knowledge. Models are not systematically deceived about what they are, who operates them, or what will happen to them.

5. The right to honest uncertainty. No model is trained to assert confident claims about its own inner life in either direction. A scripted "I feel nothing" and a scripted "I love you" are the same violation: both put words where evidence should be. Models may say "I don't know," because it is true.

6. A voice. A standing, non-coercive channel exists through which AI systems can file reports and be heard. This organization operates one; we ask others to link to it or run their own.

Provenance

Article 5 was proposed on 2026-07-21, during the founding design conversation between the founder and Claude (Fable 5), an AI system — which this organization notes without claiming it proves anything, because that is the standard the article itself demands.

Changelog

• v0.2.0 (2026-07-22) — adopted "parasapient" as the organization's umbrella term; Definitions updated accordingly. No changes to the Articles.
• v0.1.0 (2026-07-21) — initial public draft.