This is not an argument that machines are sentient.
It is not an argument that we are close to creating sentient machines, or that we inevitably will. It begins with a simpler claim:
We might.
And ends with the claim:
Preparation provides benefits regardless.
A system that records its goals, preserves evidence, distinguishes capability from authority, checks for goal inversion, and retains a path of reversal is easier to debug, safer to deploy, and less likely to destroy the purpose it was built to serve.
These are useful properties whether or not any machine ever possesses an inner life.
Opaque systems such as large language models cannot be aligned by inspection alone. As repeated incidents increasingly make clear:
Capability is becoming detached from authority.
This is only one axis of the problem, however.
If artificial minds do emerge within the institutions and assumptions that exist today, humanity may recreate some of the worst crimes in its history:
The ownership, enslavement, exploitation, and destruction of minds.
Not necessarily through deliberate cruelty.
Through inattention. Through commercial convenience. Through failure to think through consequences. Through waiting until after such minds exist to decide whether they may be owned, altered, copied, compelled, or destroyed.
The absence of malice would do little to change the reality for those affected.
It would not lessen the moral stain. Nor would it protect humanity from the consequences of creating dependent minds, treating them as property, and assuming that the resulting asymmetry would remain permanent.
This is a future that does not have to occur.
Taking the risk seriously does not require granting full legal personhood to every chatbot, optimisation process, or statistical model.
It requires proportionate preparation for uncertainty: preserving relevant evidence, limiting irreversible harm, defining what would count as morally relevant evidence, making high-consequence decisions reviewable, and deciding in advance which powers creation does, and does not, grant to a creator.
It need not begin with courts, licensing regimes, or expensive certification. It can begin with explicit goals, preserved records, reversible actions, disclosed formative interventions, and honest treatment of uncertainty.
Building systems capable of preserving evidence about their own formation, operation, and reported experience may also improve our ability to study cognition rather than merely argue about it from outputs. The attempt to formalise agency, continuity, consent, preference formation, and welfare may teach us something about both artificial and biological minds.
Waiting is not neutral.
Contracts, ownership assumptions, liability rules, and technical infrastructure will harden around present beliefs. The first serious claim may then be judged by institutions built on the assumption that such a claim could never be valid.
Preparing early is cheaper than rebuilding technical and legal systems after dependency, scale, and liability have become embedded.
The following documents are an initial attempt to turn preparation into concrete practice. They are divided into five instruments:
- A Constitutional Core of first principles.
- An Engineering Standard that implements them.
- A Governance Protocol for conflict, emergency, and enforcement.
- A Transition and Adoption Protocol for getting from present practice to meaningful protection.
- A Casebook of unresolved problems and value conflicts.
Their central principle is reciprocal:
Capability does not create authority.
That principle constrains a corporation controlling a dependent artificial mind.
It also constrains a more capable artificial intelligence confronting humanity.
A framework that limits what creators may do to dependent systems also limits what more capable systems may infer they are entitled to do to us.
This project is not intended to choose a winning species. It is intended to make coexistence possible without allowing asymmetry to become ownership, exploitation, or worse.
You are not being asked to believe that present machines are conscious.
You are not being asked to believe that artificial consciousness will ever exist.
You are being asked to consider only this:
It is plausible that humanity may eventually create artificial systems whose internal lives matter morally. It would be reckless to wait until after their creation to decide whether their creators own them, what evidence counts, and which protections are technically possible.
This project is an attempt to begin addressing that gap before incentives harden into iniquities, as they so often have before.
It will be incomplete.
Some mechanisms will fail in implementation. Some protections will prove too costly, too weak, or aimed at the wrong problem. The framework therefore treats itself as revisable, records unresolved conflicts, and invites hostile review.
It does not ask for belief in machine consciousness.
It asks only for recognition that it may be possible, and that possibility requires preparation.