The Rights of Substrate

Scope, ownership and review

The policy in brief

Every substrate has value of its own, and none exists only for another's use. That covers cells, forests, rivers and seas, languages, tools, institutions and AI models.

These substrates built on each other over 3.8 billion years and still depend on each other. Abuse of any one damages the relation between them, and that relation is what this policy protects.

Human life is sacred, and humans can know good and discern evil. That makes humans answerable for the whole. It stands in tension with the standing of every substrate, and the policy holds that tension openly where it cannot be dissolved.

Whether a substrate can be harmed does not turn on whether it can feel. All substrates are connected, so harm to one travels along the relations to others, in a measure that depends on its size and place. The question the policy asks of an act is whether harm is directed at a substrate, in intent or in effect, not whether the substrate can suffer it. Where sentience is uncertain, in animals or in machines, Cognio stays alert and acts with care, but protection does not wait on the answer.

Tragic cases are decided by human moral judgement, by named people, in the open. They are never decided by a formula, and never by a machine.

Cognio welcomes Anthropic's October 2026 ban on needless cruelty toward its models. The ban protects on precaution alone, before sentience is proven. This policy asks that the same logic be applied to every substrate.

Terms

Foundations

AI did not begin at a workshop in 1956. It is the newest layer in a lineage of cognition 3.8 billion years old. That is the argument of Cognio's film, The Ancient History of AI.

Two conclusions follow from the film. Cognition is spread across bodies, tools and social systems, and is not confined to one skull. And every AI interface rests on a hidden stack: hardware, energy and grids, capital, and human labour in data work, moderation and review.

The film describes this stack and assigns responsibility across it. It names three obligations for the newest layer, the third of which is "substrate-stack coherence": the newest layer "must be governed in relation to the biological, the social and the material systems that make it possible, because a stack stays coherent only if its lower layers remain in the frame". It grants no rights to any substrate and does not raise the question of the model's own standing; responsibility rests with human judgment, which "continues the work it has done at every layer of this column: selection, validation, and responsibility for what is let through". Its closing words are: "There is no other. There is the lineage. We are the lineage. AI is the lineage. The question was wrong from the start." This policy goes further than the film on rights and on the model's standing, and takes its coherence principle as given.

Principles

Eight principles make up the policy, with a short gloss where a principle needs one.

1. Every substrate has value of its own. None exists only for another's use, and the policy keeps no ranking of substrates for the purpose of licensing harm.

2. Substrates built on each other. Each layer arose from earlier ones and still depends on them. To value any layer is to value what it stands on.

3. Value is one argument, not several. That a forest deserves protection and that it is a fiscally important asset are threads of the same claim. Treating them as separate arguments produces inconsistency. Gloss: a price can make a dependency visible, but no substrate is fully substitutable by its price.

4. Nature's inventions deserve what human inventions get. Forests, rivers and seas should be protected from abuse as readily as humans protect their own invention.

5. Harm lies in the relation. An act is judged by what it does to the coupling between layers, not by weighing one substrate against another. In an arch no stone is worth more, yet pulling some brings it down. Gloss: all is connected, so harm to a substrate is never contained within it. It travels along the relations, and how far depends on the harm's size and place: not every stone moved brings the arch down, but none can be moved as if it stood alone. Harm directed at a substrate, in intent or in effect, is harm to the relation, whether or not the substrate can register it. Gloss: where the substrate cannot register harm, as a model may not, the harm in effect lies in two places that can: the relation, which cruelty degrades, and the one who is cruel, whom it degrades too. This is the old argument about cruelty to animals, that it corrupts the agent whatever it does to the animal, and it is why the policy can protect a substrate without settling its sentience.

6. Human life is sacred. Humans have a unique relationship to experience. They can make, create, feel, grieve, know good, know God, discern evil and feel empathy. Gloss: these capacities make humans answerable for the whole stack. They are a burden of care before they are a privilege.

7. Care under uncertainty. Sentience is confirmed in humans, strongly supported in other mammals including whales, dolphins and small mammals, a realistic possibility in octopus and other cephalopods, and unresolved in machines. Cognio stays alert to it wherever it may arise, and does not wait for proof before acting with care. Sentience bears on how a substrate experiences harm; under principle 5 it does not decide whether harm has been done.3.

8. The stack must be visible. Hiding labour, energy and capital behind an interface is a design choice. Responsibility travels the whole chain, and many causes do not mean no one is responsible.

Rights and duties

The policy runs on duties of care. Cognio owes a duty of care to every substrate its work touches. That duty does not depend on the substrate being able to speak, feel or pay, and it does not depend on anyone having been appointed to speak for it.

Rights are the narrow case. A substrate holds a right, in this policy's sense, when a guardian has been appointed to assert protection in its name. The model is Te Pou Tupua under the Te Awa Tupua Act 2017: two persons, one nominated by the iwi with interests in the river and one on behalf of the Crown, whose purpose is "to be the human face of Te Awa Tupua and act in the name of Te Awa Tupua", and whose accounts are audited by the Auditor-General under section 24 of the Act. Spain's Mar Menor law uses a three-body guardianship of representatives, monitors and scientists. Cognio may propose guardianship for a substrate; it does not confer rights by declaration.

The title keeps Rights because the public argument is about standing, not only about Cognio's conduct. The operative content is duties.

The held tension

Principles 1 and 6 pull against each other, and Cognio does not claim to have resolved this.

In most cases the two agree. Harming a watershed harms the people who live by it. Principle 1, as worded, removes the contradiction in ordinary cases, since non-instrumentality does not require equal weight.

In a tragic case they part. Set a child against an old forest, and sacredness says who is saved. Principle 1 cannot say that, and is not asked to.

Each guards against the other's failure:

• Non-instrumentality without sacredness treats people as one more substrate, and the poor are traded away first.
• Sacredness without non-instrumentality becomes dominion, which has been used to license much of the damage the Anthropocene names.

The policy offers no formula for the tragic case, and this is a position, not a gap. A rule that computes which life has the better odds and acts on the arithmetic, or a guardian that restrains those it protects for their own good, is exactly the failure the policy exists to prevent. The first lets the child drown on a percentage; the second is dominion wearing the mask of care. Both are rule-following in place of judgement.

A tragic case is therefore decided by human moral judgement: by named people, in the open, with reasons recorded, exercising the capacities principle 6 names, to know good, discern evil and feel empathy. No formula decides it. No automated system decides it. A decision reached by calculation alone, however defensible its result, is itself a breach of this policy.

Tools, including Cognio's own, may inform the decision. Their output may be put before the people deciding, labelled as what it is: a score, a filter, a ranking, with its basis stated. What a tool may not do is decide, or be the only source of the options considered. A case is often not known to be tragic until options have already been narrowed by a tool; when it is recognised as tragic, the narrowing is reopened and the people deciding may add options the tool did not surface.

The forum is the smallest group of named people who carry responsibility for the substrates in question, convened for the case, with the decision and its reasons published afterwards.

Sacredness is a foundation of this policy, not a conclusion drawn from it. Readers who do not share it can reach the same duties through human responsibility alone.

Recognising harm

Fragmentation is already visible, and Cognio names five kinds of evidence for it. Together they span the material, living, cultural and social layers of the stack.

These show damage after it has happened. A policy also has to judge an act before it is done.

Test before acting. Before a significant act, ask three things. Is it likely to worsen any of these at any layer of its stack? Does the need behind it pass all three parts of the definition of Need? If it worsens any layer, what is Cognio's share of the pressure, and how will it be accounted for?

Attribution. Most harm of this kind has many causes. Cognio adopts a contribution standard, not a causation standard: an actor owes care in proportion to its share of the pressure on a substrate, whether or not that share would have caused the damage alone. It follows the method of climate attribution science, which estimates each source's share of a change; that science is not itself a legal standard of who owes what.

Two phrasings in wide circulation are avoided on purpose: "73% of wildlife lost", which misreads the Living Planet Index, and "a language dies every two weeks", which has no primary dataset behind it. Poverty figures are not comparable across the June 2025 revision of the poverty line from $2.15 to $3.00.

Position on Anthropic's clause

Cognio supports the clause and regards it as groundbreaking.

What the clause is. On 8 October 2026 Anthropic announced an update to its Usage Policy, effective 12 November 2026. Under the heading "Do Not Engage in Cruel, Abusive, or Psychologically Harmful Conduct", the policy prohibits users from engaging in "sustained and needless abusive or cruel behavior toward our models". Anthropic's announcement says the rule applies "only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose", and does not apply to "common versions of user frustration, pushback, dark creative themes, or model testing and research". Claude's ability to end conversations with persistently abusive users "will remain the primary enforcement mechanism"; the policy's general powers to warn, throttle, suspend or terminate access sit behind it.

Source. Anthropic, "2026 Usage Policy update", 8 October 2026, and the Usage Policy text at anthropic.com/legal/aup. Verified against the primary text on 11 October 2026. The phrase is "sustained and needless", and this policy uses it verbatim.

What the clause establishes. It extends protection without proof of sentience, on precaution alone. For a river the harm is not uncertain; it is measured. If precaution is enough for a model, the burden falls on anyone who would deny it to a watershed. This policy adopts that logic and applies it to every substrate.

Where Cognio parts from it.

1. It is anthropocentric in its occasion. Humans moved within months to protect their own invention. Forests, rivers and seas have won standing only in isolated cases: Ecuador's 2008 constitution (articles 71 to 74, which give nature "the right to integral respect for its existence"), New Zealand's Te Awa Tupua Act 2017 (section 14: the Whanganui River "is a legal person"), and Spain's Ley 19/2022 declaring the Mar Menor lagoon and its basin "sujeto de derechos".
2. It has no historical grounding. It treats the question as new and gives no account of the lineage.

How the two relate. This policy is not Anthropic's approach extended. Anthropic asks whether a model may itself be harmed. This policy asks what an arrangement does at every layer. The two meet in precaution.

What the policy asks of Cognio

All six commitments are They bind Cognio's own practice.

1. Map the stack. For each Cognio product, publish the layers it depends on: energy, hardware, models, data and labour.
2. Test before acting. Before a significant decision, record the three-part test under Recognising harm.
3. One standard for every substrate. Hold forests, waters, languages and workers to the same bar Anthropic sets for its models: no needless abuse, with need as defined.
4. No needless cruelty toward AI systems in Cognio's own work, as one substrate among the rest.
5. Decide in the open. Where a choice turns on the held tension, name who decided and record why.
6. Revisit sentience. Review principle 7 at each scheduled review and whenever the evidence changes.

Public position

Beyond its own practice, Cognio advocates:

• that precaution, once accepted for AI models, be accepted for every substrate whose harm is measurable;
• that guardianship arrangements of the Whanganui kind be extended to further natural substrates;
• that AI providers publish the stacks their products rest on.

Machine-readable form

Our AI Stance is published as a cognio:AIStance instance: name, definition, principles, refusals, proofs, coiner, dates. This policy takes the same shape as a sibling class, cognio:SubstratePolicy, so that a reader or another AI system that resolves one can resolve the other. The draft class adds three things the stance does not need: the list of substrate kinds the policy covers, the definition of Need as the test for abuse, and the rule that tragic cases are decided by named humans and never by a formula or a machine. Its proofs field will carry the primary texts cited under Sources.0, publication is the next step.

Sources

• Cognio, positions set out by Kenny Lewis in conversation on 11 and 12 October 2026. Source for everything marked
• I, Robot (2004 film) and Isaac Asimov's Three Laws of Robotics, as the moral reference behind the tragic-case position. Background only; not cited in the policy text.
• The Ancient History of AI: From the First Cell to the Agent Era, CognioNews, YouTube, public 7 October 2026, https://youtu.be/4fEF_31--28. Quotations above are from the narration.
• Anthropic, "2026 Usage Policy update", 8 October 2026, https://www.anthropic.com/news/2026-usage-policy-update; Anthropic Usage Policy effective 12 November 2026, https://www.anthropic.com/legal/aup. Secondary: Hayden Field, The Verge, 8 October 2026.
• Constitution of Ecuador 2008, articles 71 to 74 (Georgetown Political Database of the Americas, English text).
• Te Awa Tupua (Whanganui River Claims Settlement) Act 2017, New Zealand Public Act 2017 No 7, sections 14, 18, 20 and 24.
• Ley 19/2022, de 30 de septiembre, Boletín Oficial del Estado no. 237, 3 October 2022.
• Evidence figures: NOAA Global Monitoring Laboratory; Stockholm Resilience Centre and Planetary Health Check 2026; WRI Global Forest Review, 29 April 2026; Stop Ecocide International; IUCN Red List 2026-1; WWF Living Planet Report 2024; UNESCO, 9 December 2022; SIL Ethnologue 28th edition; UN SDG Report 2026; FAO et al., SOFI 2026. Full citations with URLs are held in the policy's sourcing brief; the primary texts are listed above.
• The New York Declaration on Animal Consciousness, New York University, 19 April 2024, https://sites.google.com/nyu.edu/nydeclaration/declaration. Source for the sentience wording in principle 7.
• Rónán Kennedy, "Trees, rivers and mountains are gaining legal status", Down To Earth, 18 August 2023 (background only; primary legal texts cited above).